{ "text": "大家好,今天我們給大家講解Voice Agent的這個適時語音識別ASR的一個原理。\n我們從頭到尾把它捋一遍,要知道它背後的原理是什麼。當然具體的這個技術實現,肯定是由對應的這個供應商已經實現了。但是我們了解原理的話,對我們理解Voice Agent是有幫助的。\n好\n那么ASR呢\n其实我们一般会说ICTT\n就是语音转文字\n但是更专业的说法可以说ASR\n并且在这个voice agent里面\n一定要用\n实施语音\n实施语音的转换\n你整段语音的转文本\n对于这个voice agent意义不大\n因为你无法做这些\n打断了这个处理\n好\n那么他不是听懂了整句在翻译\n而是把音频声音切成一帧帧的逐帧\n拆这个什么音再把它串成\n这个再串成这个字\n是这么来的\n然后第一步把声音变成模型能吃的东西\n比如说声波到一串数字\n每秒采集16000个点\n气压变化变成一串整数\n第二部分呢\n就是切成一帧一帧的\n每帧20秒\n这么短的时间内\n声音特征基本不变\n那就是20毫秒\n20毫秒切变成一帧\n所以我们说之前说了一个40毫秒\n其实就是两帧嘛\n第三个是每帧算出频谱指纹\n不看波形\n看频率成分\n人耳听的就是这个\n其实这个是声学的里面\n最基本的原理\n这个呢\n其实\n我们不用过去顺究\n我们简单了解就行了\n为本要转成平谱了\n同一个R\n你说和我说\n大声说和小声说\n波形完全不同\n但平谱成分分布是相似的\n换成平谱模型\n要认的东西就稳定多了\n说白就说\n你说R\n我说R\n是吧\n那两个声音大小\n可能是不同的\n波形是不同的\n所以用通过波形是不行的\n要把它换成平谱\n模型才能更好的识别\n好\n第二步就是升学模型逐针猜\n这一针是什么音\n比如说你好\n对吧\n为例\n那模型对每一针独立给出一个猜测\n他不知道这个字\n只知道这一针最像什么\n那每一针猜测\n比如理\n这个是空格\n空的空白的\n理理理\n空白好好好\n空白\n灰色是空白的符号\n意思是这一针没有内容输出\n有了它\n模型就不用纠结\n一个音到底跨几帧\n说了长就多图几帧\n同样字交给下一步收拾\n所以说我们可以看到\n可能是这么个情况\n这个是每一帧的情况\n每一帧\n每一帧我们说了20毫秒\n第三步就解码\n合并重复去掉空白\n比如说我们可以看到\n有理理理好好好\n那肯定是合并成理好就行了\n然后再把这些空白的地方\n再把它去掉就变成理好\n其实就是用户就说了个理好\n好好好\n你好\n那么连续的这个\n空白的哈\n这段没人说话\n所以很多流色ASR\n根本不用单独跑一个VED\n直接读升学模型空白输出就行\n判断依据识别\n本身天然一致\n那你说白了就是有\n你这个针里面有这么多空白的符号\n那肯定就是你没说话\n对吧\n那这个就相当于做了一个VED的判断\n就是你没说话\n这就是断句原理那张图里第二步的最常见的实验方式之一\n空白符嘛\n那不就断句了吗\n我就认为你没说话\n比如说你这里面\n如果你这空白的地方\n并且时间呢\n这个比如说\n这个超过300毫秒了\n那我们认为\n你这句话说完了\n这个意思\n那么为什么牛市的识别会边缩边改呢\n对吧\n为什么他会边说边改\n比如他听到0.3秒\n离\n听到0.7秒\n离好\n听到1.2秒\n离好吗\n那么在这个时候\n他静音不是很久\n也就是说\n他是低于300毫秒的\n其实这句话就没说完\n离好这时候也没说完\n但是到这个时候\n可能后面的静音\n已经超过300毫秒了\n就final了\n就静音够久定稿了\n我们来看右边这段话\n我们看右边的解释\n流失模型只看过去看不到未来\n比如像理解的模型whisper\n拿到整段音频\n可以前后互相印证\n牛4必须在\n这个前后互相印证是什么意思\n就是他有一段完整的音频\n他已经把一个逐字稿全部练完了\n那你当然知道前后有什么\n语言他会把它改掉的\n但牛4必须在\n我想聊聊乒乓的时候就输出\n而后面是球还是球拍\n根本没有发生\n他也不知道你是聊乒乓球\n还是聊乒乓球猪\n还是聊乒乓球技术\n还是聊乒乓球拍\n对吧\n那这个牛市的就需要解决这种问题\n所以他边拆边改,改的过程呢\n就是这个\n就是我们前面说的这两段对吧\n就是in term\n那么确定不了的就是final的\n那么前面这一部分他就逐字就是在屏幕上了\n是吧\n用户才能看到他的听\n那我们经常看到这个\n比如说你在用豆包的经常就说你\n你问他问题干什么\n就是他\n或者说别的这种\n语音的这种工具\n你会发现他的字在逐步在出是吧\n就是这个意思\n他在编出还帮你改\n中文尤其是没有词的编辑\n同音字扎堆\n对吧\n他同音字特别多\n全靠上下文区分\n而上下文恰恰是流是最缺的部分\n这也就是中文识别了\n既然用阿里云\n而不是用deepgram\n那么deepgram\n他是对英文字是特别好\n包括Yenion Live这些\n所以你中文的话你还是用阿里云、火山、腾讯\n像这些是吧\n他肯定中文制作的好一些\n好\n那我们上面讲了学会之后怎么用\n推理\n那他是怎么学会的\n他为什么能识别呢\n为什么他能识别出来了\n背后的原理是什么\n第一个就是为他配好的\n对的数据\n训练数据标注预料\n有东北画,对吧,天津画\n等等,比如说上海画\n那你有这么多这个\n方言,普通话\n等等这些画\n你录给他\n他去训练这个数据\n这个是和其他的\n这个模型是一样的,你得有训练数据\n让他自己找规律,然后就训练嘛,就推理嘛,这个是为数据嘛,然后这个就是推理\n反向\n反向传播掉参数\n反复对照这几百万条\n没有人告诉他\n你这个音长什么样\n规律是他自己从数据里总结出来的\n学出来是两样东西\n升线模型\n认这一针是什么音\n语言模型\n同音字里挑哪个更适合上下文\n说白了\n就你的这个模型\n给他的数据越多\n训练的越\n找的规律越好\n对吧\n你的参数越多\n那当然你的这个识别\n可能越准\n那么学会了就可以放话\n对吧\n和推理\n那这一段\n从没听过的录音\n换一个没听过的人来说\n他也认得出\n这就是上边言那个逐针猜的\n用的模型\n逐针猜嘛\n对吧\n他要逐针猜\n比如说你说\n我要去打乒乓\n他会猜出乒乓球\n乒乓球球拍乒乓球桌乒乓球技术\n这属于这个\n我们说这个ASR用的对应的\n背后用的这种声学模型\n对吧\n或者说这个语言模型\n它就具备了这个能力了\n它也能猜\n能猜你\n说什么\n就像教小孩真智你不会给他\n讲字形\n学理论就是反复指的说\n这是礼\n是吧\n礼子多了\n他以后就会见过了字体语言的粗\n模型学的不是规则是见的多\n所以训练时\n见的少的东西\n专业术语人民方言他有容易出错\n这也是供应商为什么提供热词表\n所以说就算你的中文呢\n当然我们用中文的这种\n模型更好对吧\n英文就用英文的模型\n是吧\n你别的语言当然也是\n这个所以说是有区别的\n这么你\n你在你的这个voice in的这个项目的时候\n你在挑选来一家供应商的时候\n那肯定是根据你的业务场景来的\n你是什么语言为主\n比如说你是粤语的\n那当然你要考虑他这个模型对粤语的识别的怎么样\n识别好你才用它\n而不是说随便拿一个provider\n就行不行了\n然后我们理解了原理之后\n你对去挑选你的ASR的对应的服务\n甚至你在做本地的\n比如我们有欧拉玛去接本地的模型\n你挑哪个模型\n那都是有帮助的\n好那么关于语音识别的原理的\n这个分享了\n我们就讲到这里\n谢谢大家观看", "language": "zh", "duration": 629.423313, "segments": [ { "start": 0.52, "end": 7.16, "text": "大家好,今天我們給大家講解Voice Agent的這個適時語音識別ASR的一個原理。", "chunk": 0, "language": "zh", "avg_logprob": -0.2731893062591553, "no_speech_prob": 9.814754044212837e-12, "compression_ratio": 1.2671755725190839 }, { "start": 7.94, "end": 22.78, "text": "我們從頭到尾把它捋一遍,要知道它背後的原理是什麼。當然具體的這個技術實現,肯定是由對應的這個供應商已經實現了。但是我們了解原理的話,對我們理解Voice Agent是有幫助的。", "chunk": 0, "language": "zh", "avg_logprob": -0.2731893062591553, "no_speech_prob": 9.814754044212837e-12, "compression_ratio": 1.2671755725190839 }, { "start": 23.3, "end": 23.8, "text": "好", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 24.32, "end": 25.86, "text": "那么ASR呢", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 26.1, "end": 28.92, "text": "其实我们一般会说ICTT", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 29.18, "end": 30.72, "text": "就是语音转文字", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 31.48, "end": 35.32, "text": "但是更专业的说法可以说ASR", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 35.58, "end": 37.88, "text": "并且在这个voice agent里面", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 38.14, "end": 39.42, "text": "一定要用", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 40.44, "end": 41.46, "text": "实施语音", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 41.72, "end": 42.5, "text": "实施语音的转换", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 42.74, "end": 45.06, "text": "你整段语音的转文本", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 45.3, "end": 47.1, "text": "对于这个voice agent意义不大", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 47.36, "end": 49.14, "text": "因为你无法做这些", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 49.66, "end": 50.94, "text": "打断了这个处理", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 51.7, "end": 51.96, "text": "好", "chunk": 0, "language": "zh", "avg_logprob": -0.25806277990341187, "no_speech_prob": 1.13844402654939e-11, "compression_ratio": 1.2161016949152543 }, { "start": 52.78, "end": 55.78, "text": "那么他不是听懂了整句在翻译", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 55.78, "end": 58.82, "text": "而是把音频声音切成一帧帧的逐帧", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 58.82, "end": 61.9, "text": "拆这个什么音再把它串成", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 61.9, "end": 65.02, "text": "这个再串成这个字", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 65.02, "end": 66.08, "text": "是这么来的", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 66.08, "end": 73.24, "text": "然后第一步把声音变成模型能吃的东西", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 73.24, "end": 76.02, "text": "比如说声波到一串数字", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 76.02, "end": 80.02, "text": "每秒采集16000个点", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 80.02, "end": 82.74, "text": "气压变化变成一串整数", "chunk": 0, "language": "zh", "avg_logprob": -0.19718616264910738, "no_speech_prob": 1.2750003310080249e-11, "compression_ratio": 1.2946428571428572 }, { "start": 82.74, "end": 84.52, "text": "第二部分呢", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 84.52, "end": 85.74, "text": "就是切成一帧一帧的", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 85.74, "end": 87.12, "text": "每帧20秒", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 87.12, "end": 88.2, "text": "这么短的时间内", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 88.2, "end": 89.74, "text": "声音特征基本不变", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 89.74, "end": 90.68, "text": "那就是20毫秒", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 90.68, "end": 92.26, "text": "20毫秒切变成一帧", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 92.26, "end": 96.72, "text": "所以我们说之前说了一个40毫秒", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 96.72, "end": 97.6, "text": "其实就是两帧嘛", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 97.6, "end": 102.88, "text": "第三个是每帧算出频谱指纹", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 102.88, "end": 103.92, "text": "不看波形", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 103.92, "end": 106.0, "text": "看频率成分", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 106.0, "end": 107.28, "text": "人耳听的就是这个", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 107.28, "end": 110.3, "text": "其实这个是声学的里面", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 110.3, "end": 111.36, "text": "最基本的原理", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 111.36, "end": 111.9, "text": "这个呢", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 111.9, "end": 112.22, "text": "其实", "chunk": 0, "language": "zh", "avg_logprob": -0.16519561869986105, "no_speech_prob": 1.3869995429960369e-11, "compression_ratio": 1.4112903225806452 }, { "start": 112.22, "end": 114.0, "text": "我们不用过去顺究", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 114.0, "end": 115.1, "text": "我们简单了解就行了", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 115.1, "end": 117.18, "text": "为本要转成平谱了", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 117.18, "end": 118.38, "text": "同一个R", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 118.38, "end": 120.26, "text": "你说和我说", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 120.26, "end": 121.58, "text": "大声说和小声说", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 121.58, "end": 122.52, "text": "波形完全不同", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 122.52, "end": 126.48, "text": "但平谱成分分布是相似的", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 126.48, "end": 127.72, "text": "换成平谱模型", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 127.72, "end": 129.44, "text": "要认的东西就稳定多了", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 129.44, "end": 130.22, "text": "说白就说", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 130.22, "end": 131.12, "text": "你说R", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 131.12, "end": 132.16, "text": "我说R", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 132.16, "end": 132.54, "text": "是吧", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 132.54, "end": 133.54, "text": "那两个声音大小", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 133.54, "end": 134.32, "text": "可能是不同的", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 134.32, "end": 135.32, "text": "波形是不同的", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 135.32, "end": 136.88, "text": "所以用通过波形是不行的", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 136.88, "end": 138.36, "text": "要把它换成平谱", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 138.36, "end": 140.96, "text": "模型才能更好的识别", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 140.96, "end": 142.06, "text": "好", "chunk": 0, "language": "zh", "avg_logprob": -0.20201404615380297, "no_speech_prob": 2.0538095529820666e-11, "compression_ratio": 1.4886363636363635 }, { "start": 142.06, "end": 145.4, "text": "第二步就是升学模型逐针猜", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 145.4, "end": 146.68, "text": "这一针是什么音", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 146.68, "end": 149.36, "text": "比如说你好", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 149.36, "end": 150.12, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 150.12, "end": 150.44, "text": "为例", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 150.44, "end": 153.26, "text": "那模型对每一针独立给出一个猜测", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 153.26, "end": 155.62, "text": "他不知道这个字", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 155.62, "end": 157.72, "text": "只知道这一针最像什么", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 157.72, "end": 159.48, "text": "那每一针猜测", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 159.48, "end": 159.98, "text": "比如理", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 159.98, "end": 160.74, "text": "这个是空格", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 160.74, "end": 162.22, "text": "空的空白的", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 162.22, "end": 163.2, "text": "理理理", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 163.2, "end": 165.2, "text": "空白好好好", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 165.2, "end": 166.06, "text": "空白", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 166.06, "end": 167.98, "text": "灰色是空白的符号", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 167.98, "end": 170.02, "text": "意思是这一针没有内容输出", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 170.02, "end": 170.46, "text": "有了它", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 170.46, "end": 171.54, "text": "模型就不用纠结", "chunk": 0, "language": "zh", "avg_logprob": -0.1840721586215421, "no_speech_prob": 1.61240829715803e-11, "compression_ratio": 1.4875 }, { "start": 171.54, "end": 173.14, "text": "一个音到底跨几帧", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 173.14, "end": 175.56, "text": "说了长就多图几帧", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 175.56, "end": 177.44, "text": "同样字交给下一步收拾", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 177.44, "end": 179.12, "text": "所以说我们可以看到", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 179.12, "end": 179.96, "text": "可能是这么个情况", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 179.96, "end": 182.32, "text": "这个是每一帧的情况", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 182.32, "end": 182.88, "text": "每一帧", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 182.88, "end": 184.52, "text": "每一帧我们说了20毫秒", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 184.52, "end": 186.36, "text": "第三步就解码", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 186.36, "end": 189.14, "text": "合并重复去掉空白", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 189.14, "end": 190.58, "text": "比如说我们可以看到", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 190.58, "end": 191.78, "text": "有理理理好好好", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 191.78, "end": 194.78, "text": "那肯定是合并成理好就行了", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 194.78, "end": 198.08, "text": "然后再把这些空白的地方", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 198.08, "end": 199.42, "text": "再把它去掉就变成理好", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 199.42, "end": 200.88, "text": "其实就是用户就说了个理好", "chunk": 0, "language": "zh", "avg_logprob": -0.15199564174263777, "no_speech_prob": 2.143085361949737e-11, "compression_ratio": 1.4858156028368794 }, { "start": 200.88, "end": 202.64, "text": "好好好", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 202.64, "end": 203.48, "text": "你好", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 205.64, "end": 207.68, "text": "那么连续的这个", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 208.48, "end": 209.56, "text": "空白的哈", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 210.08, "end": 211.12, "text": "这段没人说话", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 211.12, "end": 212.6, "text": "所以很多流色ASR", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 212.6, "end": 214.4, "text": "根本不用单独跑一个VED", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 215.04, "end": 217.6, "text": "直接读升学模型空白输出就行", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 217.6, "end": 218.76, "text": "判断依据识别", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 218.76, "end": 219.76, "text": "本身天然一致", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 220.24, "end": 221.6, "text": "那你说白了就是有", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 221.6, "end": 223.92, "text": "你这个针里面有这么多空白的符号", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 223.92, "end": 225.6, "text": "那肯定就是你没说话", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 225.6, "end": 226.32, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 226.48, "end": 229.0, "text": "那这个就相当于做了一个VED的判断", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 229.56, "end": 230.68, "text": "就是你没说话", "chunk": 0, "language": "zh", "avg_logprob": -0.32948119919021407, "no_speech_prob": 1.4280924533205397e-11, "compression_ratio": 1.3370786516853932 }, { "start": 231.88, "end": 236.32, "text": "这就是断句原理那张图里第二步的最常见的实验方式之一", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 237.52, "end": 238.36, "text": "空白符嘛", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 238.36, "end": 239.24, "text": "那不就断句了吗", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 239.24, "end": 240.2, "text": "我就认为你没说话", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 241.44, "end": 242.92, "text": "比如说你这里面", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 242.92, "end": 244.36, "text": "如果你这空白的地方", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 244.64, "end": 245.6, "text": "并且时间呢", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 246.56, "end": 247.56, "text": "这个比如说", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 248.52, "end": 249.92, "text": "这个超过300毫秒了", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 249.92, "end": 250.84, "text": "那我们认为", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 251.52, "end": 252.6, "text": "你这句话说完了", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 253.04, "end": 253.56, "text": "这个意思", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 255.56, "end": 259.48, "text": "那么为什么牛市的识别会边缩边改呢", "chunk": 0, "language": "zh", "avg_logprob": -0.33687330392690806, "no_speech_prob": 1.2833792188693405e-11, "compression_ratio": 1.32 }, { "start": 259.48, "end": 260.84, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 260.84, "end": 262.76, "text": "为什么他会边说边改", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 262.76, "end": 266.2, "text": "比如他听到0.3秒", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 266.2, "end": 267.52, "text": "离", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 267.52, "end": 270.44, "text": "听到0.7秒", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 270.44, "end": 270.9, "text": "离好", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 270.9, "end": 272.88, "text": "听到1.2秒", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 272.88, "end": 273.44, "text": "离好吗", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 273.44, "end": 276.48, "text": "那么在这个时候", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 276.48, "end": 277.68, "text": "他静音不是很久", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 277.68, "end": 278.22, "text": "也就是说", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 278.22, "end": 279.48, "text": "他是低于300毫秒的", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 280.16, "end": 283.16, "text": "其实这句话就没说完", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 283.16, "end": 285.42, "text": "离好这时候也没说完", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 285.42, "end": 287.94, "text": "但是到这个时候", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 287.94, "end": 289.02, "text": "可能后面的静音", "chunk": 0, "language": "zh", "avg_logprob": -0.22305668377485433, "no_speech_prob": 1.3131290325929523e-11, "compression_ratio": 1.3705583756345177 }, { "start": 289.02, "end": 290.26, "text": "已经超过300毫秒了", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 290.86, "end": 291.66, "text": "就final了", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 291.78, "end": 293.38, "text": "就静音够久定稿了", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 296.26, "end": 298.98, "text": "我们来看右边这段话", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 308.22, "end": 309.42, "text": "我们看右边的解释", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 309.86, "end": 312.46, "text": "流失模型只看过去看不到未来", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 313.42, "end": 315.1, "text": "比如像理解的模型whisper", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 315.38, "end": 316.38, "text": "拿到整段音频", "chunk": 0, "language": "zh", "avg_logprob": -0.38911985509535846, "no_speech_prob": 1.4264196594726553e-11, "compression_ratio": 1.0939226519337018 }, { "start": 316.38, "end": 319.12, "text": "可以前后互相印证", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 319.12, "end": 320.22, "text": "牛4必须在", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 320.92, "end": 322.22, "text": "这个前后互相印证是什么意思", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 322.22, "end": 324.48, "text": "就是他有一段完整的音频", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 324.48, "end": 327.52, "text": "他已经把一个逐字稿全部练完了", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 327.72, "end": 329.72, "text": "那你当然知道前后有什么", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 330.42, "end": 332.38, "text": "语言他会把它改掉的", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 332.38, "end": 333.82, "text": "但牛4必须在", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 334.18, "end": 336.88, "text": "我想聊聊乒乓的时候就输出", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 337.18, "end": 339.04, "text": "而后面是球还是球拍", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 339.04, "end": 339.92, "text": "根本没有发生", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 340.18, "end": 341.74, "text": "他也不知道你是聊乒乓球", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 341.74, "end": 343.04, "text": "还是聊乒乓球猪", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 343.24, "end": 344.38, "text": "还是聊乒乓球技术", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 344.38, "end": 345.48, "text": "还是聊乒乓球拍", "chunk": 0, "language": "zh", "avg_logprob": -0.26768274364357225, "no_speech_prob": 1.94500561490063e-11, "compression_ratio": 1.5243445692883895 }, { "start": 346.0, "end": 346.5, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 347.78, "end": 349.84, "text": "那这个牛市的就需要解决这种问题", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 350.34, "end": 352.64, "text": "所以他边拆边改,改的过程呢", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 354.18, "end": 354.96, "text": "就是这个", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 355.72, "end": 357.76, "text": "就是我们前面说的这两段对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 361.1, "end": 361.86, "text": "就是in term", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 362.38, "end": 363.92, "text": "那么确定不了的就是final的", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 364.94, "end": 368.26, "text": "那么前面这一部分他就逐字就是在屏幕上了", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 369.04, "end": 369.54, "text": "是吧", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 369.54, "end": 370.82, "text": "用户才能看到他的听", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 371.34, "end": 372.62, "text": "那我们经常看到这个", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 372.88, "end": 374.92, "text": "比如说你在用豆包的经常就说你", "chunk": 0, "language": "zh", "avg_logprob": -0.23849317762586805, "no_speech_prob": 1.494414834934954e-11, "compression_ratio": 1.335907335907336 }, { "start": 374.92, "end": 376.52, "text": "你问他问题干什么", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 376.52, "end": 377.12, "text": "就是他", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 377.4, "end": 378.6, "text": "或者说别的这种", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 379.68, "end": 380.48, "text": "语音的这种工具", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 380.48, "end": 382.52, "text": "你会发现他的字在逐步在出是吧", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 382.76, "end": 383.68, "text": "就是这个意思", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 384.52, "end": 386.2, "text": "他在编出还帮你改", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 387.12, "end": 390.0, "text": "中文尤其是没有词的编辑", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 390.0, "end": 391.08, "text": "同音字扎堆", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 391.28, "end": 391.56, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 391.56, "end": 392.72, "text": "他同音字特别多", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 392.92, "end": 394.24, "text": "全靠上下文区分", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 394.36, "end": 396.44, "text": "而上下文恰恰是流是最缺的部分", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 396.6, "end": 398.16, "text": "这也就是中文识别了", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 398.28, "end": 399.16, "text": "既然用阿里云", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 399.16, "end": 400.32, "text": "而不是用deepgram", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 400.4, "end": 401.08, "text": "那么deepgram", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 401.08, "end": 402.36, "text": "他是对英文字是特别好", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 402.36, "end": 403.68, "text": "包括Yenion Live这些", "chunk": 0, "language": "zh", "avg_logprob": -0.3884963576858108, "no_speech_prob": 2.3177012392627816e-11, "compression_ratio": 1.3240740740740742 }, { "start": 403.68, "end": 408.08, "text": "所以你中文的话你还是用阿里云、火山、腾讯", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 408.08, "end": 409.04, "text": "像这些是吧", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 409.04, "end": 410.32, "text": "他肯定中文制作的好一些", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 410.32, "end": 411.76, "text": "好", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 411.76, "end": 418.4, "text": "那我们上面讲了学会之后怎么用", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 418.4, "end": 419.08, "text": "推理", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 419.08, "end": 421.44, "text": "那他是怎么学会的", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 421.44, "end": 424.7, "text": "他为什么能识别呢", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 424.7, "end": 427.16, "text": "为什么他能识别出来了", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 427.16, "end": 428.64, "text": "背后的原理是什么", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 428.64, "end": 431.04, "text": "第一个就是为他配好的", "chunk": 0, "language": "zh", "avg_logprob": -0.3322303142022649, "no_speech_prob": 1.1633459820470371e-11, "compression_ratio": 1.3472222222222223 }, { "start": 431.04, "end": 432.46, "text": "对的数据", "chunk": 0, "language": "zh", "avg_logprob": -0.12092553014340608, "no_speech_prob": 7.2404669065884164e-12, "compression_ratio": 0.8372093023255814 }, { "start": 432.46, "end": 434.04, "text": "训练数据标注预料", "chunk": 0, "language": "zh", "avg_logprob": -0.12092553014340608, "no_speech_prob": 7.2404669065884164e-12, "compression_ratio": 0.8372093023255814 }, { "start": 449.192, "end": 451.492, "text": "有东北画,对吧,天津画", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 451.752, "end": 453.792, "text": "等等,比如说上海画", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 454.312, "end": 456.352, "text": "那你有这么多这个", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 456.872, "end": 458.412, "text": "方言,普通话", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 458.912, "end": 459.952, "text": "等等这些画", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 460.192, "end": 461.232, "text": "你录给他", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 462.512, "end": 464.032, "text": "他去训练这个数据", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 465.072, "end": 466.352, "text": "这个是和其他的", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 466.852, "end": 468.912, "text": "这个模型是一样的,你得有训练数据", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 470.692, "end": 475.312, "text": "让他自己找规律,然后就训练嘛,就推理嘛,这个是为数据嘛,然后这个就是推理", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 475.552, "end": 476.072, "text": "反向", "chunk": 1, "language": "zh", "avg_logprob": -0.21355932995789034, "no_speech_prob": 1.442537322232651e-11, "compression_ratio": 1.4008810572687225 }, { "start": 476.072, "end": 479.192, "text": "反向传播掉参数", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 479.192, "end": 481.452, "text": "反复对照这几百万条", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 481.452, "end": 483.472, "text": "没有人告诉他", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 483.472, "end": 484.652, "text": "你这个音长什么样", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 484.652, "end": 487.372, "text": "规律是他自己从数据里总结出来的", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 487.372, "end": 488.612, "text": "学出来是两样东西", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 488.612, "end": 489.432, "text": "升线模型", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 489.432, "end": 491.312, "text": "认这一针是什么音", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 491.312, "end": 493.352, "text": "语言模型", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 493.352, "end": 495.492, "text": "同音字里挑哪个更适合上下文", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 495.492, "end": 497.212, "text": "说白了", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 497.212, "end": 498.512, "text": "就你的这个模型", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 498.512, "end": 500.452, "text": "给他的数据越多", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 500.452, "end": 501.252, "text": "训练的越", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 501.252, "end": 503.992, "text": "找的规律越好", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 503.992, "end": 504.592, "text": "对吧", "chunk": 1, "language": "zh", "avg_logprob": -0.1786103911747206, "no_speech_prob": 1.4064084966070034e-11, "compression_ratio": 1.311023622047244 }, { "start": 504.592, "end": 506.192, "text": "你的参数越多", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 506.192, "end": 507.872, "text": "那当然你的这个识别", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 507.872, "end": 508.972, "text": "可能越准", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 508.972, "end": 513.112, "text": "那么学会了就可以放话", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 513.112, "end": 513.532, "text": "对吧", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 513.532, "end": 515.612, "text": "和推理", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 515.612, "end": 517.272, "text": "那这一段", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 517.272, "end": 519.132, "text": "从没听过的录音", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 519.132, "end": 521.092, "text": "换一个没听过的人来说", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 521.092, "end": 521.852, "text": "他也认得出", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 521.852, "end": 525.412, "text": "这就是上边言那个逐针猜的", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 525.412, "end": 525.992, "text": "用的模型", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 525.992, "end": 527.452, "text": "逐针猜嘛", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 527.452, "end": 528.612, "text": "对吧", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 528.612, "end": 529.572, "text": "他要逐针猜", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 529.572, "end": 530.852, "text": "比如说你说", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 530.852, "end": 532.372, "text": "我要去打乒乓", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 532.372, "end": 533.792, "text": "他会猜出乒乓球", "chunk": 1, "language": "zh", "avg_logprob": -0.21048224938882365, "no_speech_prob": 1.4338575465844272e-11, "compression_ratio": 1.3404255319148937 }, { "start": 533.792, "end": 536.872, "text": "乒乓球球拍乒乓球桌乒乓球技术", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 537.632, "end": 538.652, "text": "这属于这个", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 539.932, "end": 541.732, "text": "我们说这个ASR用的对应的", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 541.992, "end": 543.512, "text": "背后用的这种声学模型", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 543.772, "end": 544.032, "text": "对吧", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 544.552, "end": 545.832, "text": "或者说这个语言模型", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 546.332, "end": 548.132, "text": "它就具备了这个能力了", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 548.392, "end": 549.152, "text": "它也能猜", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 549.412, "end": 550.172, "text": "能猜你", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 550.432, "end": 550.952, "text": "说什么", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 551.712, "end": 554.532, "text": "就像教小孩真智你不会给他", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 554.792, "end": 555.812, "text": "讲字形", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 556.312, "end": 558.112, "text": "学理论就是反复指的说", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 558.372, "end": 559.132, "text": "这是礼", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 559.912, "end": 560.412, "text": "是吧", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 560.672, "end": 561.952, "text": "礼子多了", "chunk": 1, "language": "zh", "avg_logprob": -0.2606331380208333, "no_speech_prob": 1.6046898185240188e-11, "compression_ratio": 1.3403361344537814 }, { "start": 561.952, "end": 564.252, "text": "他以后就会见过了字体语言的粗", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 564.252, "end": 566.052, "text": "模型学的不是规则是见的多", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 566.052, "end": 567.072, "text": "所以训练时", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 567.332, "end": 568.872, "text": "见的少的东西", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 568.872, "end": 571.432, "text": "专业术语人民方言他有容易出错", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 571.932, "end": 574.492, "text": "这也是供应商为什么提供热词表", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 575.012, "end": 576.552, "text": "所以说就算你的中文呢", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 576.552, "end": 578.852, "text": "当然我们用中文的这种", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 579.352, "end": 580.392, "text": "模型更好对吧", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 580.392, "end": 581.672, "text": "英文就用英文的模型", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 582.432, "end": 582.952, "text": "是吧", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 583.192, "end": 584.732, "text": "你别的语言当然也是", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 585.752, "end": 587.032, "text": "这个所以说是有区别的", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 587.032, "end": 587.552, "text": "这么你", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 587.812, "end": 589.592, "text": "你在你的这个voice in的这个项目的时候", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 589.592, "end": 591.652, "text": "你在挑选来一家供应商的时候", "chunk": 1, "language": "zh", "avg_logprob": -0.2272216839026351, "no_speech_prob": 1.8627946407612228e-11, "compression_ratio": 1.4681528662420382 }, { "start": 591.652, "end": 593.652, "text": "那肯定是根据你的业务场景来的", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 593.652, "end": 595.492, "text": "你是什么语言为主", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 595.492, "end": 597.112, "text": "比如说你是粤语的", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 597.112, "end": 601.672, "text": "那当然你要考虑他这个模型对粤语的识别的怎么样", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 601.672, "end": 603.012, "text": "识别好你才用它", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 603.012, "end": 605.572, "text": "而不是说随便拿一个provider", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 605.572, "end": 606.912, "text": "就行不行了", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 606.912, "end": 609.652, "text": "然后我们理解了原理之后", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 609.652, "end": 614.772, "text": "你对去挑选你的ASR的对应的服务", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 614.772, "end": 616.292, "text": "甚至你在做本地的", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 616.292, "end": 618.012, "text": "比如我们有欧拉玛去接本地的模型", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 618.012, "end": 619.152, "text": "你挑哪个模型", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 619.152, "end": 620.732, "text": "那都是有帮助的", "chunk": 1, "language": "zh", "avg_logprob": -0.14004667551239575, "no_speech_prob": 1.966200292857767e-11, "compression_ratio": 1.3355048859934853 }, { "start": 620.732, "end": 624.232, "text": "好那么关于语音识别的原理的", "chunk": 1, "language": "zh", "avg_logprob": -0.2471031733921596, "no_speech_prob": 7.887345290780168e-12, "compression_ratio": 0.9117647058823529 }, { "start": 624.232, "end": 626.192, "text": "这个分享了", "chunk": 1, "language": "zh", "avg_logprob": -0.2471031733921596, "no_speech_prob": 7.887345290780168e-12, "compression_ratio": 0.9117647058823529 }, { "start": 626.192, "end": 627.932, "text": "我们就讲到这里", "chunk": 1, "language": "zh", "avg_logprob": -0.2471031733921596, "no_speech_prob": 7.887345290780168e-12, "compression_ratio": 0.9117647058823529 }, { "start": 627.932, "end": 628.732, "text": "谢谢大家观看", "chunk": 1, "language": "zh", "avg_logprob": -0.2471031733921596, "no_speech_prob": 7.887345290780168e-12, "compression_ratio": 0.9117647058823529 } ], "segmented_transcription": { "status": "SEGMENTED_TRANSCRIPT_OK", "created_at": "2026-09-05T15:30:32", "manifest": "/Users/sagawa/AI_WORK/video_notes/url/20260905_153004__05VoiceAgent理论基础-实时语音识别ASR原理/segmented_work/chunk_manifest.json", "audio": "/Users/sagawa/AI_WORK/video_notes/url/20260905_153004__05VoiceAgent理论基础-实时语音识别ASR原理/audio_16k_mono.wav", "audio_sha256": "468d906feb1652d749cc45638e4cacc4039324c714d8921913358b70742a7db8", "audio_duration": 629.423313, "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": 304, "cleaned_segment_count": 304, "removed_hallucination_segment_count": 0, "final_segment_end": 628.732, "global_repeat_run": 1, "global_repeat_text": "大家好,今天我們給大家講解Voice Agent的這個適時語音識別ASR的一個原理。", "tail_repeat_run": 1, "tail_repeat_text": "见的少的东西", "hallucination_region_count": 0, "hallucination_regions": [], "qc_ok": true, "qc_errors": [] } }