{ "text": "好的,现在市面上有很多优秀的语音识别模型。我们有Whisper,有Parachute,还有Canary。\n如果你只需要简单的转录文本,可能会觉得这已经是解决好的问题了。\n但最大的挑战在于,如果你想为大语言模型或正在构建的应用获取上下文,仅靠转录文本是不够的。\n你通常还需要时间出来知道某句话是何时说的。\n但往往更重要的是,你需要知道到底是谁说的。\n也就是说,如果你正在为四个人互相辩论的播客制作转录文本,\n你需要知道谁说了什么,谁持哪方观点等。\n如果你想做会议纪要之类的东西,道理也是一样的。\n你需要知道谁同意了哪项行动,\n尤其是现在我们要构建智能体来处理这些事物,\n能够提取引语并准确归属于对应的人\n而不是只有一份所有人都在同一通话中逐字说出的转录文本\n就像只有一个人说话是那样\n那么一直都有办法做到这一点,对吧?\n在我之前的一些视频中\n我就展示过用Whisper配合当时的一些开源工具来做到这一点\n但这一直是一场猫鼠游戏\n看你能做到多准确\n而且你还得面对如何处理\n由不同提供商制作的多个模型等挑战\n这就引出了NVIDIA刚刚发布的东西\n这是一个专门针对该问题的新模型\n这就是NEMU闯三说话人分离\n它是一个开放权重模型\n它采用的许可证允许商业使用\n凭借这种许可证\n你当然可以做大多数你想做的事\n这里酷的一点是\n它只有大约1亿参数量\n所以只需要4GB显存的GPU就能运行\n除此之外\n当我们谈论准确率时\n它可以处理多达8位说话人\n过去大多数模型的上限\n大概是3到4位说话人\n而它不仅能在大家清晰说话时\n处理8位说话人\n即使人们互相打断说话\n它也能做得相当不错\n此外\n它的设计初衷是与你正在使用的\n任何语音识别模型配合使用\n所以如果你有一个特别喜欢的模型,比如PireKit,你就可以基本上使用这个模型。\n稍后我会展示一个演示,我们将原始播客录音转换为转录文本。\n每一行都归属于对应的说话人。\n但首先,让我介绍一下这个模型的背景。\nVDII将其作为Nemotron语音模型系列的一部分推出。\n在这个频道上,我谈过Nemotron大语言模型的,\n但我认为很多人没有意识到\n这一系列还包含很多语音模型\n在识别方面\n你有PowerKit、Canary和Nemotron ASR\n它们共同支持43种语言的流逝和批量转路\n它们还有一个叫Magpie的TTS模型\n支持多语言并具备语音定制等功能\n此外,它们还有一些进行翻译\n甚至全双公语音到语音处理的模型\n比如PersonaPlex\n直到最近\n他们的说话人分离选项\n叫做Sortformer\n最多支持四位说话人\n有一段时间\n他实际上相当受欢迎\n在很多方面\n你可以把这个新模型\n看作是对他的直接替换\n而且好得多\n我们知道\n人们已在生产环境中\n使用这些模型\n因为旧版Sortformer模型\n仅在8月就获得了\n超过30万次下载\n虽然NVIDIA没有直接说明\n但在我看来\n他们正在系统地改进这些模型\n以构建完整的Nemotron语音模型套件\n这完全说得通\n因为如果你看看对语音的需求\n无论是单用户与智能体交互\n智能体获取YouTube视频等多模态内容\n等等\n以便在研究和执行任务时提取信息作为上下文\n需求都在不断增加\n好的\n如果你是新手\n说话人分离到底是什么\n最简单的说法是\n语音识别告诉你说了什么\n而说话人分离告诉你谁说的\n当我们把它们结合起来\n就得到了所谓的带说话人信息的转录文本\n我认为大家凭直觉都知道\n这非常重要\n如果你在转录会议\n有人问了一个问题\n然后有些人回答是\n有些人回答否\n你大概希望你的智能体知道\n谁将会跟进\n谁不会采取行动\n所有这些对于智能体确定\n接下来应采取何种行动的上下文\n都至关重要\n那么它会自动知道每个说话人的名字吗\n不\n它基本上只会说说话人\n一说话人二\n说话人三\n说话人七等等\n随着处理过程进行\n但这并不难\n你可以写一些代码将其替换掉\n一旦你告诉模型谁在说话\n你就可以将转入文本\n转换为包含真实性能的版本\n现在这方面一直最大的挑战之一是\n人们会互相打断\n对吧\n即使他们并不想故意打断\n当有人在会议通话中提问时\n往往会有两三个人同时回答\n旧版说话人分离系统面临的挑战在于\n他们会因此变得混乱\n他们不擅长区分不同的声音\n尤其是在声音重叠时\n我想说\n我认为没有任何模型能在这件事上做到完美\n即使是人类\n这也可能是一项艰难的任务\n但这些模型的评分方式\n是使用所谓的说话人分离错误率\n即D.E.R\n它基本上是将三种错误相加\n模型遗漏的语音\n模型误判为非语音的内容\n以及将语音错误归属于他人的次数\n显然这里的分数越低越好\n你可以看到NVIDIA的最新模型\n不仅击败了竞争对手\n还大幅超越了他们自己的前代模型\n这里需要理解的另一个概念\n是流逝与离线的区别\n在离线模式下\n模型可以查看整个录音\n包括后面的内容\n而在流逝模式下\n它必须在音频到达时做出决定\n所以对于已经录制好的播客等\n你总是希望使用离线模式\n对于任何实时字幕或语音智能体\n你必须使用流逝模式\n好的\n那么Nemotron 3说话人分离\n到底有什么新特点\n它大约有1亿参数量\n不包括ASR部分\n运行它的最低GPU要求\n约为4GB\n所以它实际上\n可以在许多小型GPU上运行\n而且应该在笔记本电脑的\nRTX显卡上\n也能跑得相当好\n我之前提到过\n它支持多达8位说话人\n另一点是它是语言无关的\n并且在流逝和离线模式下\n都能工作\n所以我认为\n最好的办法是直接进入演示\n看看它实际上是如何运作的\n好的\n进入演示\n让我先展示一下设置方式\n首先\n感谢NVIDIA赞助这里的计算资源\n我使用DGX Spark\n来实际运行这里的三个语音模型\n让我带大家过一遍设置架构图\n你在这里看到的是NextGS应用\n它在我的Mac上本地运行\n然后通过Tail Scale调用后端道DGX Spark。\n你可以看到,我在上面有一個包含NVIDIA NEMO的Docker容器。\nVDI NEMO既提供我們正在討論的說話人分離模型,\n也提供ParaKit ASR模型和Nimetron 3.5多语言 ASR模型。\n为了使用这些,我设置了一个Fast API应用。\n这是一种实现此类功能的简单方法。\n如果我要大规模部署,我可能会选择Dynamo或Triton来提供服务。\n但对于我想要的功能,目前的设置运行得非常完美。\n如果我们进入这个应用,你可以看到我基本上可以直接使用它们的演示。\n我可以选择想要使用哪个AS2模型\n我打算用ParaKit\n因为它在处理多说话人方面\n通常表现更好\n我可以开始转录\n你可以看到字面上只需要三四秒钟\n它就处理并转录了这里\n一分三十多分钟的音频\n由于我之前在摆弄这个并设置了一些名字\n它实际上记住了这里这个人的名字\n让我们听一下\n看看当多个说话人同时说话时会发生什么\n为了庆祝Nemotron三说话人分离的发布\n我们邀请了另外七位开放文本转语音声音加入派对\n我是Chris NVIDIA的Nemotron声音\n大家好谢谢邀请\n我是Jane来自QTI的Pocket声音\n很高兴见到你们Chris和Jane\n我是Serena Quinn三的声音\n我是来自Cocorro的Adam\n只有8200万参数量\n我还是入选了嘉宾名单\n你好你好\n我是来自Resemble AI的Chatterbox\n我带来了富有表现力的能量\n嗨\n我是来自Nari Labs的John Dyer声音\n我们这里已经聚集了很棒的人群\n大家好\n最近怎么样\n我是来自Maya Research的Maya E声音\n好了\n现在我们有八位合成开放声音\n做满了场子\n让我们来测试一下Nemotron三说话人分离\n好的\n新的说话人分离模型\n可以在转路时标记我们的每个声音\n但它能在我们所有人一起说话时\n做到这一点吗\n让我们来看看\n即使我们互相打断\n它也能叫出我们俩的名字\n逐针标记\n谁在何时说话\n好的\n正如你在这里看到的\n它在转录每一个声音方面\n做得很好\n如果我们往下看看\n可以看到每个声音都被转录了\n我想我的UI在这方面\n稍微有点滞后\n但你完全可以感受到\n它是如何组装起来的\n实际上\n如果我们放大一点\n来自QTI的声音\n很高兴见到你们\nChris和Jen\n我是Serena\nQueen3的声音\n你可以很快看到\n我们基本上可以组装出一个能转录这些声音的东西\n如果我们想在这里输入他们的名字\n它会便利并为我们重命名所有人\n现在如果我们想看看实际模型返回的是什么\n你可以看到这里是模型返回的原始Jason\n我们可以看到这里实际使用的是什么\n我们可以看到不同的说话人被标记等\n然后这个软件基本上将speaker三重写为Adam等\n便利这个过程\n做这样的事情\n很酷的一点是\n我们可以很快\n获得统计数据\n比如每个人\n在那里说了多久\n说了多少次\n有多少次发言\n如果我们想在这里\n放入更长的内容\n这会非常有用\n好的\n这里酷的是\n我们可以处理长文件\n所以我基本上\n放入了\nAll in 播客的\n一集完整节目\n他们采访了黄仁勋\n你可以看到\n他基本上已经处理完了\n我们得到了输出\n我命名了这里的实际说话人\n所以我们可以看到\n说话人依次出现\n现在我们有了每个人\n说了什么的完整转录文本\n这里酷的是我们可以查看统计数据\n我们可以看到他们各自的说话时长\n我们可以看到\n果然\n显然\n黄仁勋是这里的嘉宾\n所以他是说话最多的\n而这一切都在148秒内完成\n所以对于这个超过一小时的长文件\n这大约花了两分半钟\n这让我们基本上可以便利\n你可以看到\n现在我们基本上可以直接阅读转录文本\n我可以将其导出为文本文件\n或者在这里导出为srt\n这里酷的是\n如果我确实想做多语言\n或其他类似的事情\n我可以切换模型\n使用这个模型\n而不是这里的paraket模型\n你可以看到\n用这个模型\n我们支持很多不同的语言\n如果你记得我做过关于这个的视频\n你实际上可以微调这个模型\n使其在其他语言上表现非常好\n他们在这里称之为适配就绪\n现在我还没有真正设置这个应用去做\n但模型能做的另一件事\n是你也可以进行流逝响应\n当然\n那样你的准确率可能会下降一点\n但你在那里仍然应该能获得相当好的结果\n所以\n如果你有兴趣做任何语音转录\n并且希望其中包含说话人分离等\n这是一个很好的设置方法\n你可以批量完成所有转路等\n最重要的是\n能够完全在本地完成所有这一切\n所以\n如果你对这方面有任何问题\n请在评论区告诉我\n我认为看到NVIDIA\n接下来在这些语音模型上\n会做什么将会非常酷\n就像我说的\n从我看到的来看\n它们不断迭代\n以改进这些模型\n并为不同的用力\n提供不同类型的模型\n总之\n一如既往\n如果你觉得视频有用\n请点赞并订阅\n我们下个视频见\n再见", "language": "zh", "duration": 789.24625, "segments": [ { "start": 0.0, "end": 7.1, "text": "好的,现在市面上有很多优秀的语音识别模型。我们有Whisper,有Parachute,还有Canary。", "chunk": 0, "language": "zh", "avg_logprob": -0.1110308453753278, "no_speech_prob": 9.223401556401889e-12, "compression_ratio": 1.220125786163522 }, { "start": 9.2, "end": 14.3, "text": "如果你只需要简单的转录文本,可能会觉得这已经是解决好的问题了。", "chunk": 0, "language": "zh", "avg_logprob": -0.1110308453753278, "no_speech_prob": 9.223401556401889e-12, "compression_ratio": 1.220125786163522 }, { "start": 14.8, "end": 23.72, "text": "但最大的挑战在于,如果你想为大语言模型或正在构建的应用获取上下文,仅靠转录文本是不够的。", "chunk": 0, "language": "zh", "avg_logprob": -0.1110308453753278, "no_speech_prob": 9.223401556401889e-12, "compression_ratio": 1.220125786163522 }, { "start": 24.22, "end": 28.1, "text": "你通常还需要时间出来知道某句话是何时说的。", "chunk": 0, "language": "zh", "avg_logprob": -0.1110308453753278, "no_speech_prob": 9.223401556401889e-12, "compression_ratio": 1.220125786163522 }, { "start": 28.1, "end": 33.0, "text": "但往往更重要的是,你需要知道到底是谁说的。", "chunk": 0, "language": "zh", "avg_logprob": -0.1237065350567853, "no_speech_prob": 1.3636162511654337e-11, "compression_ratio": 1.3431734317343174 }, { "start": 34.32, "end": 39.62, "text": "也就是说,如果你正在为四个人互相辩论的播客制作转录文本,", "chunk": 0, "language": "zh", "avg_logprob": -0.1237065350567853, "no_speech_prob": 1.3636162511654337e-11, "compression_ratio": 1.3431734317343174 }, { "start": 39.94, "end": 43.94, "text": "你需要知道谁说了什么,谁持哪方观点等。", "chunk": 0, "language": "zh", "avg_logprob": -0.1237065350567853, "no_speech_prob": 1.3636162511654337e-11, "compression_ratio": 1.3431734317343174 }, { "start": 43.94, "end": 47.68, "text": "如果你想做会议纪要之类的东西,道理也是一样的。", "chunk": 0, "language": "zh", "avg_logprob": -0.1237065350567853, "no_speech_prob": 1.3636162511654337e-11, "compression_ratio": 1.3431734317343174 }, { "start": 48.14, "end": 50.7, "text": "你需要知道谁同意了哪项行动,", "chunk": 0, "language": "zh", "avg_logprob": -0.1237065350567853, "no_speech_prob": 1.3636162511654337e-11, "compression_ratio": 1.3431734317343174 }, { "start": 51.28, "end": 54.9, "text": "尤其是现在我们要构建智能体来处理这些事物,", "chunk": 0, "language": "zh", "avg_logprob": -0.1237065350567853, "no_speech_prob": 1.3636162511654337e-11, "compression_ratio": 1.3431734317343174 }, { "start": 54.9, "end": 58.68, "text": "能够提取引语并准确归属于对应的人", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 58.68, "end": 65.22, "text": "而不是只有一份所有人都在同一通话中逐字说出的转录文本", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 65.22, "end": 67.36, "text": "就像只有一个人说话是那样", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 67.36, "end": 70.0, "text": "那么一直都有办法做到这一点,对吧?", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 70.72, "end": 72.08, "text": "在我之前的一些视频中", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 72.08, "end": 76.56, "text": "我就展示过用Whisper配合当时的一些开源工具来做到这一点", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 76.56, "end": 78.74, "text": "但这一直是一场猫鼠游戏", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 78.74, "end": 80.12, "text": "看你能做到多准确", "chunk": 0, "language": "zh", "avg_logprob": -0.09461426734924316, "no_speech_prob": 1.1306598018956393e-11, "compression_ratio": 1.2961672473867596 }, { "start": 80.12, "end": 83.18, "text": "而且你还得面对如何处理", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 83.18, "end": 86.14, "text": "由不同提供商制作的多个模型等挑战", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 86.14, "end": 90.74, "text": "这就引出了NVIDIA刚刚发布的东西", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 90.74, "end": 93.7, "text": "这是一个专门针对该问题的新模型", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 93.7, "end": 96.56, "text": "这就是NEMU闯三说话人分离", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 96.56, "end": 98.24, "text": "它是一个开放权重模型", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 98.24, "end": 100.94, "text": "它采用的许可证允许商业使用", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 100.94, "end": 102.22, "text": "凭借这种许可证", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 102.22, "end": 104.48, "text": "你当然可以做大多数你想做的事", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 104.48, "end": 105.92, "text": "这里酷的一点是", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 105.92, "end": 107.88, "text": "它只有大约1亿参数量", "chunk": 0, "language": "zh", "avg_logprob": -0.09645654504949397, "no_speech_prob": 1.358369926957037e-11, "compression_ratio": 1.2673267326732673 }, { "start": 107.88, "end": 111.76, "text": "所以只需要4GB显存的GPU就能运行", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 111.76, "end": 112.66, "text": "除此之外", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 112.66, "end": 114.66, "text": "当我们谈论准确率时", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 114.66, "end": 116.92, "text": "它可以处理多达8位说话人", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 116.92, "end": 120.12, "text": "过去大多数模型的上限", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 120.12, "end": 122.38, "text": "大概是3到4位说话人", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 122.38, "end": 125.34, "text": "而它不仅能在大家清晰说话时", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 125.34, "end": 126.78, "text": "处理8位说话人", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 126.78, "end": 129.26, "text": "即使人们互相打断说话", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 129.26, "end": 130.8, "text": "它也能做得相当不错", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 130.8, "end": 131.7, "text": "此外", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 131.7, "end": 134.12, "text": "它的设计初衷是与你正在使用的", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 134.12, "end": 136.52, "text": "任何语音识别模型配合使用", "chunk": 0, "language": "zh", "avg_logprob": -0.07662676061902728, "no_speech_prob": 1.0536456256093896e-11, "compression_ratio": 1.3028169014084507 }, { "start": 136.52, "end": 142.04, "text": "所以如果你有一个特别喜欢的模型,比如PireKit,你就可以基本上使用这个模型。", "chunk": 0, "language": "zh", "avg_logprob": -0.09639582951863607, "no_speech_prob": 1.3422976272159381e-11, "compression_ratio": 1.2774193548387096 }, { "start": 142.34, "end": 147.28, "text": "稍后我会展示一个演示,我们将原始播客录音转换为转录文本。", "chunk": 0, "language": "zh", "avg_logprob": -0.09639582951863607, "no_speech_prob": 1.3422976272159381e-11, "compression_ratio": 1.2774193548387096 }, { "start": 147.44, "end": 150.34, "text": "每一行都归属于对应的说话人。", "chunk": 0, "language": "zh", "avg_logprob": -0.09639582951863607, "no_speech_prob": 1.3422976272159381e-11, "compression_ratio": 1.2774193548387096 }, { "start": 150.52, "end": 153.84, "text": "但首先,让我介绍一下这个模型的背景。", "chunk": 0, "language": "zh", "avg_logprob": -0.09639582951863607, "no_speech_prob": 1.3422976272159381e-11, "compression_ratio": 1.2774193548387096 }, { "start": 154.26, "end": 158.36, "text": "VDII将其作为Nemotron语音模型系列的一部分推出。", "chunk": 0, "language": "zh", "avg_logprob": -0.09639582951863607, "no_speech_prob": 1.3422976272159381e-11, "compression_ratio": 1.2774193548387096 }, { "start": 158.82, "end": 162.36, "text": "在这个频道上,我谈过Nemotron大语言模型的,", "chunk": 0, "language": "zh", "avg_logprob": -0.09639582951863607, "no_speech_prob": 1.3422976272159381e-11, "compression_ratio": 1.2774193548387096 }, { "start": 162.36, "end": 164.86, "text": "但我认为很多人没有意识到", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 164.86, "end": 167.84, "text": "这一系列还包含很多语音模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 167.84, "end": 169.08, "text": "在识别方面", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 169.08, "end": 172.3, "text": "你有PowerKit、Canary和Nemotron ASR", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 172.3, "end": 179.06, "text": "它们共同支持43种语言的流逝和批量转路", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 179.06, "end": 182.22, "text": "它们还有一个叫Magpie的TTS模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 182.22, "end": 185.42, "text": "支持多语言并具备语音定制等功能", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 185.42, "end": 187.66, "text": "此外,它们还有一些进行翻译", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 187.66, "end": 190.38, "text": "甚至全双公语音到语音处理的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10050747670283934, "no_speech_prob": 1.1706318206461397e-11, "compression_ratio": 1.196551724137931 }, { "start": 190.38, "end": 192.48, "text": "比如PersonaPlex", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 192.48, "end": 193.56, "text": "直到最近", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 193.56, "end": 195.5, "text": "他们的说话人分离选项", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 195.5, "end": 196.98, "text": "叫做Sortformer", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 196.98, "end": 198.92, "text": "最多支持四位说话人", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 198.92, "end": 199.78, "text": "有一段时间", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 199.78, "end": 201.86, "text": "他实际上相当受欢迎", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 201.86, "end": 203.3, "text": "在很多方面", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 203.3, "end": 204.88, "text": "你可以把这个新模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 204.88, "end": 206.88, "text": "看作是对他的直接替换", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 206.88, "end": 207.8, "text": "而且好得多", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 207.8, "end": 208.56, "text": "我们知道", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 208.56, "end": 210.1, "text": "人们已在生产环境中", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 210.1, "end": 211.18, "text": "使用这些模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 211.18, "end": 213.1, "text": "因为旧版Sortformer模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 213.1, "end": 214.58, "text": "仅在8月就获得了", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 214.58, "end": 216.06, "text": "超过30万次下载", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 216.06, "end": 218.5, "text": "虽然NVIDIA没有直接说明", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 218.5, "end": 219.62, "text": "但在我看来", "chunk": 0, "language": "zh", "avg_logprob": -0.11223171212080586, "no_speech_prob": 1.3048434728546443e-11, "compression_ratio": 1.1923076923076923 }, { "start": 219.62, "end": 222.42, "text": "他们正在系统地改进这些模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 222.42, "end": 226.08, "text": "以构建完整的Nemotron语音模型套件", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 226.08, "end": 228.82, "text": "这完全说得通", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 228.82, "end": 231.72, "text": "因为如果你看看对语音的需求", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 231.72, "end": 236.1, "text": "无论是单用户与智能体交互", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 236.1, "end": 241.98, "text": "智能体获取YouTube视频等多模态内容", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 241.98, "end": 243.74, "text": "等等", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 243.74, "end": 247.86, "text": "以便在研究和执行任务时提取信息作为上下文", "chunk": 0, "language": "zh", "avg_logprob": -0.11190070343017579, "no_speech_prob": 1.1619782393224032e-11, "compression_ratio": 1.1383399209486167 }, { "start": 247.86, "end": 249.76, "text": "需求都在不断增加", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 249.76, "end": 250.12, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 250.12, "end": 251.58, "text": "如果你是新手", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 251.58, "end": 253.48, "text": "说话人分离到底是什么", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 253.48, "end": 254.98, "text": "最简单的说法是", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 254.98, "end": 257.28, "text": "语音识别告诉你说了什么", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 257.28, "end": 260.06, "text": "而说话人分离告诉你谁说的", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 260.06, "end": 263.22, "text": "当我们把它们结合起来", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 263.22, "end": 266.72, "text": "就得到了所谓的带说话人信息的转录文本", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 266.72, "end": 269.06, "text": "我认为大家凭直觉都知道", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 269.06, "end": 270.1, "text": "这非常重要", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 270.1, "end": 271.86, "text": "如果你在转录会议", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 271.86, "end": 273.16, "text": "有人问了一个问题", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 273.16, "end": 274.62, "text": "然后有些人回答是", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 274.62, "end": 276.24, "text": "有些人回答否", "chunk": 0, "language": "zh", "avg_logprob": -0.10300330626658905, "no_speech_prob": 1.2513887495679832e-11, "compression_ratio": 1.407942238267148 }, { "start": 276.24, "end": 278.82, "text": "你大概希望你的智能体知道", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 278.82, "end": 280.28, "text": "谁将会跟进", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 280.28, "end": 282.04, "text": "谁不会采取行动", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 282.04, "end": 284.2, "text": "所有这些对于智能体确定", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 284.2, "end": 287.12, "text": "接下来应采取何种行动的上下文", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 287.12, "end": 288.14, "text": "都至关重要", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 288.14, "end": 290.52, "text": "那么它会自动知道每个说话人的名字吗", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 290.52, "end": 291.34, "text": "不", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 291.34, "end": 293.24, "text": "它基本上只会说说话人", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 293.24, "end": 295.1, "text": "一说话人二", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 295.1, "end": 296.28, "text": "说话人三", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 296.28, "end": 297.82, "text": "说话人七等等", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 297.82, "end": 299.64, "text": "随着处理过程进行", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 299.64, "end": 301.2, "text": "但这并不难", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 301.2, "end": 303.32, "text": "你可以写一些代码将其替换掉", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 303.32, "end": 305.46, "text": "一旦你告诉模型谁在说话", "chunk": 0, "language": "zh", "avg_logprob": -0.11096656753356199, "no_speech_prob": 1.4801550610177294e-11, "compression_ratio": 1.4105263157894736 }, { "start": 305.46, "end": 306.84, "text": "你就可以将转入文本", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 306.84, "end": 308.0, "text": "转换为包含真实性能的版本", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 308.0, "end": 310.4, "text": "现在这方面一直最大的挑战之一是", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 310.4, "end": 312.26, "text": "人们会互相打断", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 312.26, "end": 312.58, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 312.58, "end": 314.58, "text": "即使他们并不想故意打断", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 314.58, "end": 317.28, "text": "当有人在会议通话中提问时", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 317.28, "end": 319.84, "text": "往往会有两三个人同时回答", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 319.84, "end": 324.84, "text": "旧版说话人分离系统面临的挑战在于", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 324.84, "end": 326.76, "text": "他们会因此变得混乱", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 326.76, "end": 329.58, "text": "他们不擅长区分不同的声音", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 329.58, "end": 331.82, "text": "尤其是在声音重叠时", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 331.82, "end": 334.06, "text": "我想说", "chunk": 0, "language": "zh", "avg_logprob": -0.1123166601341891, "no_speech_prob": 1.3900204771932767e-11, "compression_ratio": 1.34375 }, { "start": 334.06, "end": 337.82, "text": "我认为没有任何模型能在这件事上做到完美", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 337.82, "end": 338.78, "text": "即使是人类", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 338.78, "end": 341.46, "text": "这也可能是一项艰难的任务", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 341.46, "end": 343.14, "text": "但这些模型的评分方式", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 343.14, "end": 346.04, "text": "是使用所谓的说话人分离错误率", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 346.04, "end": 347.4, "text": "即D.E.R", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 347.4, "end": 349.98, "text": "它基本上是将三种错误相加", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 349.98, "end": 352.0, "text": "模型遗漏的语音", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 352.0, "end": 354.46, "text": "模型误判为非语音的内容", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 354.46, "end": 358.18, "text": "以及将语音错误归属于他人的次数", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 358.18, "end": 360.78, "text": "显然这里的分数越低越好", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 360.78, "end": 363.64, "text": "你可以看到NVIDIA的最新模型", "chunk": 0, "language": "zh", "avg_logprob": -0.09585204069641814, "no_speech_prob": 1.4124049152464124e-11, "compression_ratio": 1.3066666666666666 }, { "start": 363.64, "end": 365.4, "text": "不仅击败了竞争对手", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 365.4, "end": 369.02, "text": "还大幅超越了他们自己的前代模型", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 369.02, "end": 371.7, "text": "这里需要理解的另一个概念", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 371.7, "end": 374.06, "text": "是流逝与离线的区别", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 374.06, "end": 375.6, "text": "在离线模式下", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 375.6, "end": 377.88, "text": "模型可以查看整个录音", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 377.88, "end": 379.78, "text": "包括后面的内容", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 379.78, "end": 381.12, "text": "而在流逝模式下", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 381.12, "end": 383.88, "text": "它必须在音频到达时做出决定", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 383.88, "end": 386.54, "text": "所以对于已经录制好的播客等", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 386.54, "end": 389.02, "text": "你总是希望使用离线模式", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 389.02, "end": 392.34, "text": "对于任何实时字幕或语音智能体", "chunk": 0, "language": "zh", "avg_logprob": -0.08109930666481577, "no_speech_prob": 1.1572645619573052e-11, "compression_ratio": 1.3309859154929577 }, { "start": 392.34, "end": 394.24, "text": "你必须使用流逝模式", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 394.24, "end": 394.82, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 394.82, "end": 397.08, "text": "那么Nemotron 3说话人分离", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 397.08, "end": 398.28, "text": "到底有什么新特点", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 398.28, "end": 400.22, "text": "它大约有1亿参数量", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 400.22, "end": 401.9, "text": "不包括ASR部分", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 401.9, "end": 403.78, "text": "运行它的最低GPU要求", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 403.78, "end": 404.68, "text": "约为4GB", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 404.68, "end": 405.96, "text": "所以它实际上", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 405.96, "end": 408.36, "text": "可以在许多小型GPU上运行", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 408.36, "end": 410.24, "text": "而且应该在笔记本电脑的", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 410.24, "end": 411.64, "text": "RTX显卡上", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 411.64, "end": 413.0, "text": "也能跑得相当好", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 413.0, "end": 414.34, "text": "我之前提到过", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 414.34, "end": 416.12, "text": "它支持多达8位说话人", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 416.12, "end": 418.3, "text": "另一点是它是语言无关的", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 418.3, "end": 420.08, "text": "并且在流逝和离线模式下", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 420.08, "end": 420.74, "text": "都能工作", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 420.74, "end": 421.74, "text": "所以我认为", "chunk": 0, "language": "zh", "avg_logprob": -0.08413103648594447, "no_speech_prob": 1.2611624418401579e-11, "compression_ratio": 1.2332361516034986 }, { "start": 421.74, "end": 424.42, "text": "最好的办法是直接进入演示", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 424.42, "end": 426.8, "text": "看看它实际上是如何运作的", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 426.8, "end": 427.66, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 427.66, "end": 429.02, "text": "进入演示", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 429.02, "end": 431.04, "text": "让我先展示一下设置方式", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 431.04, "end": 431.46, "text": "首先", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 431.46, "end": 434.6, "text": "感谢NVIDIA赞助这里的计算资源", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 434.6, "end": 436.54, "text": "我使用DGX Spark", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 436.54, "end": 439.58, "text": "来实际运行这里的三个语音模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 439.58, "end": 442.54, "text": "让我带大家过一遍设置架构图", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 442.54, "end": 446.6, "text": "你在这里看到的是NextGS应用", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 446.6, "end": 448.8, "text": "它在我的Mac上本地运行", "chunk": 0, "language": "zh", "avg_logprob": -0.11040007046290806, "no_speech_prob": 1.1481075638808402e-11, "compression_ratio": 1.2065217391304348 }, { "start": 448.8, "end": 454.0, "text": "然后通过Tail Scale调用后端道DGX Spark。", "chunk": 0, "language": "zh", "avg_logprob": -0.2842778932480585, "no_speech_prob": 1.5784996573731114e-11, "compression_ratio": 0.8448275862068966 }, { "start": 455.695, "end": 460.995, "text": "你可以看到,我在上面有一個包含NVIDIA NEMO的Docker容器。", "chunk": 1, "language": "zh", "avg_logprob": -0.2159466579042632, "no_speech_prob": 1.4088164662640068e-11, "compression_ratio": 1.0845588235294117 }, { "start": 461.495, "end": 466.095, "text": "VDI NEMO既提供我們正在討論的說話人分離模型,", "chunk": 1, "language": "zh", "avg_logprob": -0.2159466579042632, "no_speech_prob": 1.4088164662640068e-11, "compression_ratio": 1.0845588235294117 }, { "start": 466.095, "end": 472.935, "text": "也提供ParaKit ASR模型和Nimetron 3.5多语言 ASR模型。", "chunk": 1, "language": "zh", "avg_logprob": -0.10933547825008244, "no_speech_prob": 9.478564634568531e-12, "compression_ratio": 1.2235649546827794 }, { "start": 473.555, "end": 477.355, "text": "为了使用这些,我设置了一个Fast API应用。", "chunk": 1, "language": "zh", "avg_logprob": -0.10933547825008244, "no_speech_prob": 9.478564634568531e-12, "compression_ratio": 1.2235649546827794 }, { "start": 477.515, "end": 479.955, "text": "这是一种实现此类功能的简单方法。", "chunk": 1, "language": "zh", "avg_logprob": -0.10933547825008244, "no_speech_prob": 9.478564634568531e-12, "compression_ratio": 1.2235649546827794 }, { "start": 480.495, "end": 485.615, "text": "如果我要大规模部署,我可能会选择Dynamo或Triton来提供服务。", "chunk": 1, "language": "zh", "avg_logprob": -0.10933547825008244, "no_speech_prob": 9.478564634568531e-12, "compression_ratio": 1.2235649546827794 }, { "start": 485.735, "end": 489.935, "text": "但对于我想要的功能,目前的设置运行得非常完美。", "chunk": 1, "language": "zh", "avg_logprob": -0.10933547825008244, "no_speech_prob": 9.478564634568531e-12, "compression_ratio": 1.2235649546827794 }, { "start": 490.295, "end": 495.095, "text": "如果我们进入这个应用,你可以看到我基本上可以直接使用它们的演示。", "chunk": 1, "language": "zh", "avg_logprob": -0.10933547825008244, "no_speech_prob": 9.478564634568531e-12, "compression_ratio": 1.2235649546827794 }, { "start": 495.095, "end": 498.615, "text": "我可以选择想要使用哪个AS2模型", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 498.615, "end": 500.295, "text": "我打算用ParaKit", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 500.295, "end": 502.515, "text": "因为它在处理多说话人方面", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 502.515, "end": 503.835, "text": "通常表现更好", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 503.835, "end": 505.015, "text": "我可以开始转录", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 505.015, "end": 508.435, "text": "你可以看到字面上只需要三四秒钟", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 508.435, "end": 510.495, "text": "它就处理并转录了这里", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 510.495, "end": 512.595, "text": "一分三十多分钟的音频", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 512.595, "end": 519.455, "text": "由于我之前在摆弄这个并设置了一些名字", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 519.455, "end": 521.975, "text": "它实际上记住了这里这个人的名字", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 521.975, "end": 524.795, "text": "让我们听一下", "chunk": 1, "language": "zh", "avg_logprob": -0.1070260464305609, "no_speech_prob": 9.556695712564789e-12, "compression_ratio": 1.2605633802816902 }, { "start": 524.795, "end": 528.555, "text": "看看当多个说话人同时说话时会发生什么", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 528.555, "end": 532.535, "text": "为了庆祝Nemotron三说话人分离的发布", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 532.535, "end": 535.895, "text": "我们邀请了另外七位开放文本转语音声音加入派对", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 535.895, "end": 538.455, "text": "我是Chris NVIDIA的Nemotron声音", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 538.455, "end": 540.235, "text": "大家好谢谢邀请", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 540.235, "end": 543.355, "text": "我是Jane来自QTI的Pocket声音", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 543.355, "end": 545.915, "text": "很高兴见到你们Chris和Jane", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 545.915, "end": 548.015, "text": "我是Serena Quinn三的声音", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 548.015, "end": 549.635, "text": "我是来自Cocorro的Adam", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 549.635, "end": 551.375, "text": "只有8200万参数量", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 551.375, "end": 552.795, "text": "我还是入选了嘉宾名单", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 552.795, "end": 553.695, "text": "你好你好", "chunk": 1, "language": "zh", "avg_logprob": -0.11973179776244368, "no_speech_prob": 1.766951689130547e-11, "compression_ratio": 1.2454545454545454 }, { "start": 553.695, "end": 555.835, "text": "我是来自Resemble AI的Chatterbox", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 555.835, "end": 557.815, "text": "我带来了富有表现力的能量", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 557.815, "end": 558.055, "text": "嗨", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 558.055, "end": 560.435, "text": "我是来自Nari Labs的John Dyer声音", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 560.435, "end": 562.175, "text": "我们这里已经聚集了很棒的人群", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 562.175, "end": 562.795, "text": "大家好", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 562.795, "end": 563.595, "text": "最近怎么样", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 563.595, "end": 566.175, "text": "我是来自Maya Research的Maya E声音", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 566.175, "end": 566.955, "text": "好了", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 566.955, "end": 568.975, "text": "现在我们有八位合成开放声音", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 568.975, "end": 569.855, "text": "做满了场子", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 569.855, "end": 572.055, "text": "让我们来测试一下Nemotron三说话人分离", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 572.055, "end": 572.255, "text": "好的", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 572.255, "end": 573.875, "text": "新的说话人分离模型", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 573.875, "end": 575.915, "text": "可以在转路时标记我们的每个声音", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 575.915, "end": 578.055, "text": "但它能在我们所有人一起说话时", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 578.055, "end": 578.775, "text": "做到这一点吗", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 578.775, "end": 579.515, "text": "让我们来看看", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 579.515, "end": 580.915, "text": "即使我们互相打断", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 580.915, "end": 582.755, "text": "它也能叫出我们俩的名字", "chunk": 1, "language": "zh", "avg_logprob": -0.10513490923283773, "no_speech_prob": 2.1958716095449304e-11, "compression_ratio": 1.371212121212121 }, { "start": 582.755, "end": 583.775, "text": "逐针标记", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 583.775, "end": 585.515, "text": "谁在何时说话", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 585.515, "end": 586.475, "text": "好的", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 586.475, "end": 588.715, "text": "正如你在这里看到的", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 588.715, "end": 591.115, "text": "它在转录每一个声音方面", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 591.115, "end": 591.935, "text": "做得很好", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 591.935, "end": 593.375, "text": "如果我们往下看看", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 593.375, "end": 595.555, "text": "可以看到每个声音都被转录了", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 595.555, "end": 597.175, "text": "我想我的UI在这方面", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 597.175, "end": 598.495, "text": "稍微有点滞后", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 598.495, "end": 599.915, "text": "但你完全可以感受到", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 599.915, "end": 601.355, "text": "它是如何组装起来的", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 601.355, "end": 602.215, "text": "实际上", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 602.215, "end": 603.415, "text": "如果我们放大一点", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 603.415, "end": 604.755, "text": "来自QTI的声音", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 604.755, "end": 605.915, "text": "很高兴见到你们", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 605.915, "end": 607.175, "text": "Chris和Jen", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 607.175, "end": 607.975, "text": "我是Serena", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 607.975, "end": 609.335, "text": "Queen3的声音", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 609.335, "end": 610.755, "text": "你可以很快看到", "chunk": 1, "language": "zh", "avg_logprob": -0.13857351197786857, "no_speech_prob": 1.7001601515520548e-11, "compression_ratio": 1.3114754098360655 }, { "start": 610.755, "end": 615.055, "text": "我们基本上可以组装出一个能转录这些声音的东西", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 615.055, "end": 617.775, "text": "如果我们想在这里输入他们的名字", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 617.775, "end": 620.815, "text": "它会便利并为我们重命名所有人", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 620.815, "end": 625.495, "text": "现在如果我们想看看实际模型返回的是什么", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 625.495, "end": 628.915, "text": "你可以看到这里是模型返回的原始Jason", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 628.915, "end": 632.655, "text": "我们可以看到这里实际使用的是什么", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 632.655, "end": 635.475, "text": "我们可以看到不同的说话人被标记等", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 635.475, "end": 640.515, "text": "然后这个软件基本上将speaker三重写为Adam等", "chunk": 1, "language": "zh", "avg_logprob": -0.10199743954103384, "no_speech_prob": 1.091457826896125e-11, "compression_ratio": 1.4456140350877194 }, { "start": 640.515, "end": 641.855, "text": "便利这个过程", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 641.855, "end": 643.835, "text": "做这样的事情", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 643.835, "end": 644.955, "text": "很酷的一点是", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 644.955, "end": 646.015, "text": "我们可以很快", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 646.015, "end": 647.055, "text": "获得统计数据", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 647.055, "end": 648.875, "text": "比如每个人", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 648.875, "end": 650.095, "text": "在那里说了多久", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 650.095, "end": 652.335, "text": "说了多少次", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 652.335, "end": 653.715, "text": "有多少次发言", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 653.715, "end": 655.675, "text": "如果我们想在这里", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 655.675, "end": 657.135, "text": "放入更长的内容", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 657.135, "end": 658.315, "text": "这会非常有用", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 658.315, "end": 658.795, "text": "好的", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 658.795, "end": 660.115, "text": "这里酷的是", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 660.115, "end": 661.875, "text": "我们可以处理长文件", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 661.875, "end": 663.335, "text": "所以我基本上", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 663.335, "end": 663.935, "text": "放入了", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 663.935, "end": 665.035, "text": "All in 播客的", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 665.035, "end": 666.195, "text": "一集完整节目", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 666.195, "end": 667.455, "text": "他们采访了黄仁勋", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 667.455, "end": 668.355, "text": "你可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 668.355, "end": 669.815, "text": "他基本上已经处理完了", "chunk": 1, "language": "zh", "avg_logprob": -0.1224022601024214, "no_speech_prob": 2.0128333028113232e-11, "compression_ratio": 1.3745704467353952 }, { "start": 669.815, "end": 672.115, "text": "我们得到了输出", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 672.115, "end": 674.435, "text": "我命名了这里的实际说话人", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 674.435, "end": 675.795, "text": "所以我们可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 675.795, "end": 677.555, "text": "说话人依次出现", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 677.555, "end": 679.095, "text": "现在我们有了每个人", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 679.095, "end": 681.075, "text": "说了什么的完整转录文本", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 681.075, "end": 683.935, "text": "这里酷的是我们可以查看统计数据", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 683.935, "end": 686.315, "text": "我们可以看到他们各自的说话时长", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 686.315, "end": 688.415, "text": "我们可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 688.415, "end": 689.415, "text": "果然", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 689.415, "end": 690.035, "text": "显然", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 690.035, "end": 691.615, "text": "黄仁勋是这里的嘉宾", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 691.615, "end": 693.615, "text": "所以他是说话最多的", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 693.615, "end": 696.395, "text": "而这一切都在148秒内完成", "chunk": 1, "language": "zh", "avg_logprob": -0.10066930730859716, "no_speech_prob": 1.4456450793398634e-11, "compression_ratio": 1.4819277108433735 }, { "start": 696.395, "end": 699.855, "text": "所以对于这个超过一小时的长文件", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 699.855, "end": 701.815, "text": "这大约花了两分半钟", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 701.815, "end": 704.415, "text": "这让我们基本上可以便利", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 704.415, "end": 705.615, "text": "你可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 705.615, "end": 708.555, "text": "现在我们基本上可以直接阅读转录文本", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 708.555, "end": 711.035, "text": "我可以将其导出为文本文件", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 711.035, "end": 712.915, "text": "或者在这里导出为srt", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 712.915, "end": 714.135, "text": "这里酷的是", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 714.135, "end": 716.595, "text": "如果我确实想做多语言", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 716.595, "end": 718.075, "text": "或其他类似的事情", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 718.075, "end": 719.695, "text": "我可以切换模型", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 719.695, "end": 721.015, "text": "使用这个模型", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 721.015, "end": 723.335, "text": "而不是这里的paraket模型", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 723.335, "end": 724.355, "text": "你可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 724.355, "end": 725.535, "text": "用这个模型", "chunk": 1, "language": "zh", "avg_logprob": -0.09703300201814938, "no_speech_prob": 1.1800498078695654e-11, "compression_ratio": 1.4548736462093863 }, { "start": 725.535, "end": 727.775, "text": "我们支持很多不同的语言", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 727.775, "end": 730.115, "text": "如果你记得我做过关于这个的视频", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 730.115, "end": 732.455, "text": "你实际上可以微调这个模型", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 732.455, "end": 735.535, "text": "使其在其他语言上表现非常好", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 735.535, "end": 737.735, "text": "他们在这里称之为适配就绪", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 737.735, "end": 740.695, "text": "现在我还没有真正设置这个应用去做", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 740.695, "end": 742.555, "text": "但模型能做的另一件事", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 742.555, "end": 744.595, "text": "是你也可以进行流逝响应", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 744.595, "end": 744.955, "text": "当然", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 744.955, "end": 747.555, "text": "那样你的准确率可能会下降一点", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 747.555, "end": 751.255, "text": "但你在那里仍然应该能获得相当好的结果", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 751.255, "end": 752.235, "text": "所以", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 752.235, "end": 754.735, "text": "如果你有兴趣做任何语音转录", "chunk": 1, "language": "zh", "avg_logprob": -0.08130152549969374, "no_speech_prob": 1.5152703744525375e-11, "compression_ratio": 1.3545454545454545 }, { "start": 754.735, "end": 757.535, "text": "并且希望其中包含说话人分离等", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 757.535, "end": 759.675, "text": "这是一个很好的设置方法", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 759.675, "end": 762.155, "text": "你可以批量完成所有转路等", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 762.155, "end": 763.055, "text": "最重要的是", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 763.055, "end": 765.335, "text": "能够完全在本地完成所有这一切", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 765.335, "end": 766.115, "text": "所以", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 766.115, "end": 767.995, "text": "如果你对这方面有任何问题", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 767.995, "end": 769.355, "text": "请在评论区告诉我", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 769.355, "end": 770.935, "text": "我认为看到NVIDIA", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 770.935, "end": 772.995, "text": "接下来在这些语音模型上", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 772.995, "end": 774.555, "text": "会做什么将会非常酷", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 774.555, "end": 775.855, "text": "就像我说的", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 775.855, "end": 777.095, "text": "从我看到的来看", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 777.095, "end": 778.015, "text": "它们不断迭代", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 778.015, "end": 779.115, "text": "以改进这些模型", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 779.115, "end": 780.695, "text": "并为不同的用力", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 780.695, "end": 782.815, "text": "提供不同类型的模型", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 782.815, "end": 783.475, "text": "总之", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 783.475, "end": 784.515, "text": "一如既往", "chunk": 1, "language": "zh", "avg_logprob": -0.10270933671431108, "no_speech_prob": 1.385195430581021e-11, "compression_ratio": 1.3294460641399417 }, { "start": 784.515, "end": 785.975, "text": "如果你觉得视频有用", "chunk": 1, "language": "zh", "avg_logprob": -0.08529777824878693, "no_speech_prob": 1.0453256316100834e-11, "compression_ratio": 0.8888888888888888 }, { "start": 785.975, "end": 787.335, "text": "请点赞并订阅", "chunk": 1, "language": "zh", "avg_logprob": -0.08529777824878693, "no_speech_prob": 1.0453256316100834e-11, "compression_ratio": 0.8888888888888888 }, { "start": 787.335, "end": 788.535, "text": "我们下个视频见", "chunk": 1, "language": "zh", "avg_logprob": -0.08529777824878693, "no_speech_prob": 1.0453256316100834e-11, "compression_ratio": 0.8888888888888888 }, { "start": 788.535, "end": 789.195, "text": "再见", "chunk": 1, "language": "zh", "avg_logprob": -0.08529777824878693, "no_speech_prob": 1.0453256316100834e-11, "compression_ratio": 0.8888888888888888 } ], "segmented_transcription": { "status": "SEGMENTED_TRANSCRIPT_OK", "created_at": "2026-09-29T17:33:29", "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260929_173259__中配_NVIDIA_Nemotron_3_说话人分离实测_8人混音精准识别与本地部署_p01_中文配音版/segmented_work/chunk_manifest.json", "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260929_173259__中配_NVIDIA_Nemotron_3_说话人分离实测_8人混音精准识别与本地部署_p01_中文配音版/audio_16k_mono.wav", "audio_sha256": "6ff760587e807966fb5a0a9640c99fe1e173b71b2afee11375cf80b39d61640d", "audio_duration": 789.24625, "model": "/Users/sagawa/AI_WORK/cache/huggingface/hub/models--mlx-community--whisper-large-v3-turbo/snapshots/a4aaeec0636e6fef84abdcbe3544cb2bf7e9f6fb", "language_hint": "zh", "condition_on_previous_text": false, "chunk_count": 2, "resumed_chunks": 0, "merged_segment_count": 350, "cleaned_segment_count": 350, "removed_hallucination_segment_count": 0, "final_segment_end": 789.195, "global_repeat_run": 1, "global_repeat_text": "好的,现在市面上有很多优秀的语音识别模型。我们有Whisper,有Parachute,还有Canary。", "tail_repeat_run": 1, "tail_repeat_text": "但模型能做的另一件事", "hallucination_region_count": 0, "hallucination_regions": [], "qc_ok": true, "qc_errors": [] } }