{
  "text": "视频还没播完,MOSVIA Real Time已经开始描述画面。\n他先认出咖啡店和走进门的一男一女,随后又跟到了柜台前的交谈。\n后面的人物继续出现,文字也继续往下涨。\n这不是等视频结束以后再做总结,模型正在跟着画面一起走。\n它是MOS智能推出的110亿参数开放全入模型,专门处理持续进入的视频流。\n普通的视频问答往往要先看完整段再统一回答。\n事实版把等待拆开了,新画面不断进来,理解和文字生成也可以同时继续。\n按照官方介绍,你可以在视频播放过程中插入问题、证据还不够时。\n它可以先保持沉默,后面的画面改变了情境,它还能补充或者修正此前的判断。\n边看边答,主动沉默和动态更新构成了这个版本最核心的区别。\n同系列还有基础版和指令版。\n基础版适合继续训练,指令版面向完整视频的深入问答。\n实施版才专门面对持续视频流。\n官方标注的256000上下文,表示一次能够参考的信息容量,\n不是固定的视频分钟数。抽针越密,对话越长,能够容纳的视频时长就会跟着变化。\n接下来是我最关心的问题,消费机显卡能不能把它跑起来。\n官方给出了4比特量化版本,按照一张R4G显卡设计,我没有这张卡,\n于是改用两张4070Ti Super让程序自动拆分模型。\n本地部署走的是一条完整的成功路径。\n仓库拉到本地以后,我创建独立环境,装好运行依赖,再下载官方量化权重。\n启动实施推理程序时,三个权重分片依次加载。\n加载完成以后,模型开始按时间接收视频。\n两张显卡也同时参与推理。\n画面继续进入,文字随之更新,终端最后显示36帧全部送达,\n这次本地部署正式跑通。\n测试输入是一段30多秒的咖啡店视频,程序每秒取一帧,\n模型先认出了门店和进场人物,随后描述男子走向柜台,\n店员递出纸杯,场景继续变化,他又跟到了后面的人物冲突。\n整段输出不是最后一次性出现,而是随着视频推进不断补充。\n我又在视频播到约10秒时插入一个问题,问现在画面里有几个人,\n他先回答两个人。\n7秒以后,柜台后的电源进入画面,他没有等我再问,\n直接把答案更新成3个人,到了26秒我再问刚才发生了什么变化。\n他先认出电源正在介绍三种杯子,随后又补充男子已经走到柜台前,\n两次提问都发生在播放过程中,后面的画面也没有被第一轮回答挡住。\n这次运行也给出了一个很直接的硬件结论,\n我用两张4070Ti Super跑完了这段视频,扣掉系统本来的占用,\n模型额外使用的显存大约13.3G,这个结果只对应当前样片和参数,\n但至少证明这套双卡配置可以完成一次本地实施推理。\n输出中间还有一段很值得看画面没有明显变化时,\n终端没有继续增加描述,后面的人物进入以后文字才重新出现。\n原始接口会把这种状态写成控制标记,真正做成应用时需要先过滤,\n对观众来说更重要的是模型知道什么时候先观察,\n而不是每一帧都重复说话。\n不过,实时不代表每个瞬间都不会漏掉,当前测试每秒只取一帧,\n视频里一个非常短的击打动作就没有被稳定识别。\n它看到了动作前后不等于看见了中间的每一刻,\n提高采样频率可能保留更多细节,也会同时增加输入和缓存压力。\n这种持续更新来自模型的输入方式,视觉信息不是等到整段结束再交给语言部分,\n每个采样画面还带着自己的时间位置,所以事件先后和状态变化能够进入同一段推理。\n官方代码除了读取本地视频还支持摄像头、屏幕和外部发送的视频帧,\n它可以成为直播辅助实验观察、流程提醒或者屏幕任务监控的视觉底座。\n模型负责持续理解和回答,真正上线是告紧规则、权限控制和人工复合仍然要由应用层补上。\n所以Mousewell Real-Time的优势很明确,视频还在发声,问题和答案就能一起向前推进。\n它的限制也同样明确,采样会漏掉短动作,长上下文会继续消耗缓存,原始输出也不是装好就能使用的监控产品。\n如果你的任务只是等视频结束以后再总结,指令版会更直接。\n如果你需要在视频进行中观察、提问等待,再根据新画面更新判断,实施版才真正对口。\n项目地址和官方资料放在评论区,感兴趣可以拿自己的视频试一遍。",
  "language": "zh",
  "duration": 258.88,
  "segments": [
    {
      "id": 0,
      "seek": 0,
      "start": 0.0,
      "end": 3.12,
      "text": "视频还没播完,MOSVIA Real Time已经开始描述画面。",
      "tokens": [
        50365,
        40656,
        39752,
        14852,
        10062,
        49993,
        14128,
        11,
        44,
        4367,
        53,
        6914,
        8467,
        6161,
        49161,
        45213,
        11673,
        237,
        3316,
        108,
        27126,
        8833,
        1543,
        50521
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 1,
      "seek": 0,
      "start": 3.12,
      "end": 7.8,
      "text": "他先认出咖啡店和走进门的一男一女,随后又跟到了柜台前的交谈。",
      "tokens": [
        50521,
        5000,
        10108,
        7422,
        97,
        7781,
        8975,
        244,
        3284,
        94,
        22469,
        12565,
        9575,
        36700,
        8259,
        101,
        1546,
        2257,
        25243,
        2257,
        17015,
        11,
        10673,
        237,
        13547,
        17047,
        9678,
        21381,
        17238,
        250,
        15433,
        8945,
        1546,
        28455,
        8897,
        230,
        1543,
        50755
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 2,
      "seek": 0,
      "start": 7.8,
      "end": 11.200000000000001,
      "text": "后面的人物继续出现,文字也继续往下涨。",
      "tokens": [
        50755,
        13547,
        8833,
        29979,
        23516,
        10115,
        100,
        10115,
        255,
        7781,
        20204,
        11,
        17174,
        22381,
        6404,
        10115,
        100,
        10115,
        255,
        29510,
        4438,
        35681,
        101,
        1543,
        50925
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 3,
      "seek": 0,
      "start": 11.200000000000001,
      "end": 15.52,
      "text": "这不是等视频结束以后再做总结,模型正在跟着画面一起走。",
      "tokens": [
        50925,
        5562,
        7296,
        10187,
        40656,
        39752,
        45641,
        44320,
        3588,
        13547,
        8623,
        10907,
        33440,
        45641,
        11,
        41908,
        39823,
        15789,
        3581,
        9678,
        20708,
        27126,
        8833,
        29567,
        9575,
        1543,
        51141
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 4,
      "seek": 0,
      "start": 15.52,
      "end": 21.2,
      "text": "它是MOS智能推出的110亿参数开放全入模型,专门处理持续进入的视频流。",
      "tokens": [
        51141,
        11284,
        1541,
        44,
        4367,
        5094,
        118,
        8225,
        33597,
        7781,
        1546,
        43435,
        1369,
        123,
        2129,
        224,
        33188,
        18937,
        12744,
        11319,
        14028,
        41908,
        39823,
        11,
        940,
        241,
        8259,
        101,
        1787,
        226,
        13876,
        17694,
        10115,
        255,
        36700,
        14028,
        1546,
        40656,
        39752,
        27854,
        1543,
        51425
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 5,
      "seek": 0,
      "start": 21.2,
      "end": 24.6,
      "text": "普通的视频问答往往要先看完整段再统一回答。",
      "tokens": [
        51425,
        29993,
        19550,
        1546,
        40656,
        39752,
        22064,
        28223,
        29510,
        29510,
        4275,
        10108,
        4200,
        14128,
        27662,
        28427,
        8623,
        10115,
        253,
        2257,
        8350,
        28223,
        1543,
        51595
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 6,
      "seek": 0,
      "start": 24.6,
      "end": 29.68,
      "text": "事实版把等待拆开了,新画面不断进来,理解和文字生成也可以同时继续。",
      "tokens": [
        51595,
        6973,
        24726,
        42096,
        16075,
        10187,
        18390,
        6852,
        228,
        18937,
        2289,
        11,
        12560,
        27126,
        8833,
        1960,
        4307,
        255,
        36700,
        6912,
        11,
        13876,
        17278,
        12565,
        17174,
        22381,
        8244,
        11336,
        47935,
        13089,
        15729,
        10115,
        100,
        10115,
        255,
        1543,
        51849
      ],
      "temperature": 0,
      "avg_logprob": -0.13174379252951984,
      "compression_ratio": 1.2915662650602409,
      "no_speech_prob": 7.634058316052794e-12,
      "source": "whisper"
    },
    {
      "id": 7,
      "seek": 2968,
      "start": 29.68,
      "end": 34.68,
      "text": "按照官方介绍,你可以在视频播放过程中插入问题、证据还不够时。",
      "tokens": [
        50365,
        26613,
        32150,
        31929,
        9249,
        30312,
        10115,
        235,
        11,
        42766,
        3581,
        40656,
        39752,
        49993,
        12744,
        16866,
        29649,
        5975,
        11673,
        240,
        14028,
        34069,
        1231,
        5233,
        223,
        26075,
        106,
        14852,
        1960,
        1787,
        253,
        15729,
        1543,
        50615
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 8,
      "seek": 2968,
      "start": 34.68,
      "end": 39.88,
      "text": "它可以先保持沉默,后面的画面改变了情境,它还能补充或者修正此前的判断。",
      "tokens": [
        50615,
        11284,
        6723,
        10108,
        24302,
        17694,
        3308,
        231,
        6173,
        246,
        11,
        13547,
        8833,
        1546,
        27126,
        8833,
        34490,
        2129,
        246,
        2289,
        10570,
        47885,
        11,
        11284,
        14852,
        8225,
        9890,
        98,
        2347,
        227,
        31148,
        7792,
        106,
        15789,
        17947,
        8945,
        1546,
        2437,
        97,
        4307,
        255,
        1543,
        50875
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 9,
      "seek": 2968,
      "start": 39.88,
      "end": 44.28,
      "text": "边看边答,主动沉默和动态更新构成了这个版本最核心的区别。",
      "tokens": [
        50875,
        40503,
        4200,
        40503,
        28223,
        11,
        13557,
        34961,
        3308,
        231,
        6173,
        246,
        12565,
        34961,
        3757,
        223,
        19002,
        12560,
        7360,
        226,
        11336,
        2289,
        15368,
        42096,
        8802,
        8661,
        8987,
        116,
        7945,
        1546,
        9937,
        118,
        18453,
        1543,
        51095
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 10,
      "seek": 2968,
      "start": 44.28,
      "end": 46.480000000000004,
      "text": "同系列还有基础版和指令版。",
      "tokens": [
        51095,
        13089,
        25368,
        43338,
        35091,
        26008,
        38114,
        222,
        42096,
        12565,
        25922,
        49061,
        42096,
        1543,
        51205
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 11,
      "seek": 2968,
      "start": 46.480000000000004,
      "end": 50.6,
      "text": "基础版适合继续训练,指令版面向完整视频的深入问答。",
      "tokens": [
        51205,
        26008,
        38114,
        222,
        42096,
        2215,
        224,
        14245,
        10115,
        100,
        10115,
        255,
        7422,
        255,
        10115,
        225,
        11,
        25922,
        49061,
        42096,
        8833,
        24282,
        14128,
        27662,
        40656,
        39752,
        1546,
        24043,
        14028,
        22064,
        28223,
        1543,
        51411
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 12,
      "seek": 2968,
      "start": 50.6,
      "end": 53.2,
      "text": "实施版才专门面对持续视频流。",
      "tokens": [
        51411,
        24726,
        4307,
        121,
        42096,
        18888,
        940,
        241,
        8259,
        101,
        8833,
        8713,
        17694,
        10115,
        255,
        40656,
        39752,
        27854,
        1543,
        51541
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 13,
      "seek": 2968,
      "start": 53.2,
      "end": 58.08,
      "text": "官方标注的256000上下文,表示一次能够参考的信息容量,",
      "tokens": [
        51541,
        31929,
        9249,
        162,
        3921,
        26432,
        1546,
        6074,
        21,
        1360,
        5708,
        4438,
        17174,
        11,
        40053,
        27505,
        8225,
        1787,
        253,
        2129,
        224,
        26504,
        1546,
        17665,
        26460,
        25750,
        26748,
        11,
        51785
      ],
      "temperature": 0,
      "avg_logprob": -0.07606139341236856,
      "compression_ratio": 1.3478260869565217,
      "no_speech_prob": 1.428204429720914e-11,
      "source": "whisper"
    },
    {
      "id": 14,
      "seek": 5808,
      "start": 58.08,
      "end": 64.67999999999999,
      "text": "不是固定的视频分钟数。抽针越密,对话越长,能够容纳的视频时长就会跟着变化。",
      "tokens": [
        50365,
        7296,
        3919,
        118,
        12088,
        1546,
        40656,
        39752,
        6627,
        50064,
        33188,
        1543,
        46022,
        25153,
        230,
        25761,
        47838,
        11,
        8713,
        21596,
        25761,
        32271,
        11,
        8225,
        1787,
        253,
        25750,
        16853,
        111,
        1546,
        40656,
        39752,
        15729,
        32271,
        3111,
        12949,
        9678,
        20708,
        2129,
        246,
        23756,
        1543,
        50695
      ],
      "temperature": 0,
      "avg_logprob": -0.13445081443430107,
      "compression_ratio": 1.162280701754386,
      "no_speech_prob": 1.4966614753086915e-11,
      "source": "whisper"
    },
    {
      "id": 15,
      "seek": 5808,
      "start": 64.67999999999999,
      "end": 68.6,
      "text": "接下来是我最关心的问题,消费机显卡能不能把它跑起来。",
      "tokens": [
        50695,
        14468,
        4438,
        6912,
        42614,
        8661,
        28053,
        7945,
        1546,
        34069,
        11,
        28837,
        18464,
        117,
        37960,
        1431,
        122,
        32681,
        8225,
        28590,
        42061,
        32585,
        9147,
        6912,
        1543,
        50891
      ],
      "temperature": 0,
      "avg_logprob": -0.13445081443430107,
      "compression_ratio": 1.162280701754386,
      "no_speech_prob": 1.4966614753086915e-11,
      "source": "whisper"
    },
    {
      "id": 16,
      "seek": 5808,
      "start": 68.6,
      "end": 73.56,
      "text": "官方给出了4比特量化版本,按照一张R4G显卡设计,我没有这张卡,",
      "tokens": [
        50891,
        31929,
        9249,
        23197,
        7781,
        2289,
        19,
        11706,
        17682,
        26748,
        23756,
        42096,
        8802,
        11,
        26613,
        32150,
        2257,
        44059,
        49,
        19,
        38,
        1431,
        122,
        32681,
        7422,
        122,
        7422,
        94,
        11,
        1654,
        17944,
        5562,
        44059,
        32681,
        11,
        51139
      ],
      "temperature": 0,
      "avg_logprob": -0.13445081443430107,
      "compression_ratio": 1.162280701754386,
      "no_speech_prob": 1.4966614753086915e-11,
      "source": "whisper"
    },
    {
      "id": 17,
      "seek": 7356,
      "start": 73.56,
      "end": 78.16,
      "text": "于是改用两张4070Ti Super让程序自动拆分模型。",
      "tokens": [
        50365,
        37732,
        1541,
        34490,
        9254,
        36257,
        44059,
        5254,
        5867,
        51,
        72,
        4548,
        33650,
        29649,
        6346,
        237,
        9722,
        34961,
        6852,
        228,
        6627,
        41908,
        39823,
        1543,
        50595
      ],
      "temperature": 0,
      "avg_logprob": -0.09501959572375661,
      "compression_ratio": 1.1702898550724639,
      "no_speech_prob": 1.4908263859525483e-11,
      "source": "whisper"
    },
    {
      "id": 18,
      "seek": 7356,
      "start": 78.16,
      "end": 81.16,
      "text": "本地部署走的是一条完整的成功路径。",
      "tokens": [
        50595,
        8802,
        10928,
        13470,
        16469,
        110,
        9575,
        24620,
        2257,
        48837,
        14128,
        27662,
        1546,
        11336,
        32311,
        24658,
        2172,
        226,
        1543,
        50745
      ],
      "temperature": 0,
      "avg_logprob": -0.09501959572375661,
      "compression_ratio": 1.1702898550724639,
      "no_speech_prob": 1.4908263859525483e-11,
      "source": "whisper"
    },
    {
      "id": 19,
      "seek": 7356,
      "start": 81.16,
      "end": 85.88,
      "text": "仓库拉到本地以后,我创建独立环境,装好运行依赖,再下载官方量化权重。",
      "tokens": [
        50745,
        1550,
        241,
        6346,
        241,
        29717,
        4511,
        8802,
        10928,
        3588,
        13547,
        11,
        1654,
        2437,
        249,
        34157,
        18637,
        105,
        24409,
        8051,
        107,
        47885,
        11,
        8083,
        227,
        2131,
        3316,
        238,
        8082,
        3254,
        251,
        5266,
        244,
        11,
        8623,
        4438,
        17819,
        121,
        31929,
        9249,
        26748,
        23756,
        4422,
        225,
        12624,
        1543,
        50981
      ],
      "temperature": 0,
      "avg_logprob": -0.09501959572375661,
      "compression_ratio": 1.1702898550724639,
      "no_speech_prob": 1.4908263859525483e-11,
      "source": "whisper"
    },
    {
      "id": 20,
      "seek": 7356,
      "start": 85.88,
      "end": 89.04,
      "text": "启动实施推理程序时,三个权重分片依次加载。",
      "tokens": [
        50981,
        2392,
        107,
        34961,
        24726,
        4307,
        121,
        33597,
        13876,
        29649,
        6346,
        237,
        15729,
        11,
        10960,
        7549,
        4422,
        225,
        12624,
        6627,
        16668,
        3254,
        251,
        9487,
        9990,
        17819,
        121,
        1543,
        51139
      ],
      "temperature": 0,
      "avg_logprob": -0.09501959572375661,
      "compression_ratio": 1.1702898550724639,
      "no_speech_prob": 1.4908263859525483e-11,
      "source": "whisper"
    },
    {
      "id": 21,
      "seek": 7356,
      "start": 89.04,
      "end": 92.16,
      "text": "加载完成以后,模型开始按时间接收视频。",
      "tokens": [
        51139,
        9990,
        17819,
        121,
        41509,
        3588,
        13547,
        11,
        41908,
        39823,
        45213,
        26613,
        44848,
        14468,
        18681,
        40656,
        39752,
        1543,
        51295
      ],
      "temperature": 0,
      "avg_logprob": -0.09501959572375661,
      "compression_ratio": 1.1702898550724639,
      "no_speech_prob": 1.4908263859525483e-11,
      "source": "whisper"
    },
    {
      "id": 22,
      "seek": 9216,
      "start": 92.16,
      "end": 94.44,
      "text": "两张显卡也同时参与推理。",
      "tokens": [
        50365,
        36257,
        44059,
        1431,
        122,
        32681,
        6404,
        13089,
        15729,
        2129,
        224,
        940,
        236,
        33597,
        13876,
        1543,
        50479
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 23,
      "seek": 9216,
      "start": 94.44,
      "end": 99.84,
      "text": "画面继续进入,文字随之更新,终端最后显示36帧全部送达,",
      "tokens": [
        50479,
        27126,
        8833,
        10115,
        100,
        10115,
        255,
        36700,
        14028,
        11,
        17174,
        22381,
        10673,
        237,
        9574,
        19002,
        12560,
        11,
        10115,
        230,
        11957,
        107,
        8661,
        13547,
        1431,
        122,
        25696,
        11309,
        4845,
        100,
        38714,
        29309,
        9830,
        122,
        11,
        50749
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 24,
      "seek": 9216,
      "start": 99.84,
      "end": 101.75999999999999,
      "text": "这次本地部署正式跑通。",
      "tokens": [
        50749,
        5562,
        9487,
        8802,
        10928,
        13470,
        16469,
        110,
        15789,
        27584,
        32585,
        19550,
        1543,
        50845
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 25,
      "seek": 9216,
      "start": 101.75999999999999,
      "end": 106.08,
      "text": "测试输入是一段30多秒的咖啡店视频,程序每秒取一帧,",
      "tokens": [
        50845,
        11038,
        233,
        5233,
        243,
        9830,
        241,
        14028,
        1541,
        2257,
        28427,
        3446,
        6392,
        20956,
        1546,
        8975,
        244,
        3284,
        94,
        22469,
        40656,
        39752,
        11,
        29649,
        6346,
        237,
        23664,
        20956,
        29436,
        2257,
        4845,
        100,
        11,
        51061
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 26,
      "seek": 9216,
      "start": 106.08,
      "end": 110.24,
      "text": "模型先认出了门店和进场人物,随后描述男子走向柜台,",
      "tokens": [
        51061,
        41908,
        39823,
        10108,
        7422,
        97,
        7781,
        2289,
        8259,
        101,
        22469,
        12565,
        36700,
        50255,
        4035,
        23516,
        11,
        10673,
        237,
        13547,
        11673,
        237,
        3316,
        108,
        25243,
        7626,
        9575,
        24282,
        17238,
        250,
        15433,
        11,
        51269
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 27,
      "seek": 9216,
      "start": 110.24,
      "end": 115.24,
      "text": "店员递出纸杯,场景继续变化,他又跟到了后面的人物冲突。",
      "tokens": [
        51269,
        22469,
        3606,
        246,
        2215,
        240,
        7781,
        16853,
        116,
        45458,
        11,
        50255,
        50218,
        10115,
        100,
        10115,
        255,
        2129,
        246,
        23756,
        11,
        5000,
        17047,
        9678,
        21381,
        13547,
        8833,
        29979,
        23516,
        5676,
        110,
        40859,
        1543,
        51519
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 28,
      "seek": 9216,
      "start": 115.24,
      "end": 120.03999999999999,
      "text": "整段输出不是最后一次性出现,而是随着视频推进不断补充。",
      "tokens": [
        51519,
        27662,
        28427,
        9830,
        241,
        7781,
        7296,
        8661,
        13547,
        27505,
        21686,
        7781,
        20204,
        11,
        11070,
        1541,
        10673,
        237,
        20708,
        40656,
        39752,
        33597,
        36700,
        1960,
        4307,
        255,
        9890,
        98,
        2347,
        227,
        1543,
        51759
      ],
      "temperature": 0,
      "avg_logprob": -0.10659976996997796,
      "compression_ratio": 1.301775147928994,
      "no_speech_prob": 1.3896770019450333e-11,
      "source": "whisper"
    },
    {
      "id": 29,
      "seek": 12004,
      "start": 120.04,
      "end": 125.08000000000001,
      "text": "我又在视频播到约10秒时插入一个问题,问现在画面里有几个人,",
      "tokens": [
        50365,
        1654,
        17047,
        3581,
        40656,
        39752,
        49993,
        4511,
        16853,
        99,
        3279,
        20956,
        15729,
        11673,
        240,
        14028,
        20182,
        34069,
        11,
        22064,
        25040,
        27126,
        8833,
        15759,
        2412,
        6336,
        254,
        7549,
        4035,
        11,
        50617
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 30,
      "seek": 12004,
      "start": 125.08000000000001,
      "end": 126.24000000000001,
      "text": "他先回答两个人。",
      "tokens": [
        50617,
        5000,
        10108,
        8350,
        28223,
        36257,
        7549,
        4035,
        1543,
        50675
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 31,
      "seek": 12004,
      "start": 126.24000000000001,
      "end": 129.84,
      "text": "7秒以后,柜台后的电源进入画面,他没有等我再问,",
      "tokens": [
        50675,
        22,
        20956,
        3588,
        13547,
        11,
        17238,
        250,
        15433,
        13547,
        1546,
        42182,
        47402,
        36700,
        14028,
        27126,
        8833,
        11,
        5000,
        17944,
        10187,
        1654,
        8623,
        22064,
        11,
        50855
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 32,
      "seek": 12004,
      "start": 129.84,
      "end": 134.32,
      "text": "直接把答案更新成3个人,到了26秒我再问刚才发生了什么变化。",
      "tokens": [
        50855,
        43297,
        16075,
        28223,
        28899,
        19002,
        12560,
        11336,
        18,
        7549,
        4035,
        11,
        21381,
        10880,
        20956,
        1654,
        8623,
        22064,
        49160,
        18888,
        28926,
        8244,
        2289,
        10440,
        2129,
        246,
        23756,
        1543,
        51079
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 33,
      "seek": 12004,
      "start": 134.32,
      "end": 139.04000000000002,
      "text": "他先认出电源正在介绍三种杯子,随后又补充男子已经走到柜台前,",
      "tokens": [
        51079,
        5000,
        10108,
        7422,
        97,
        7781,
        42182,
        47402,
        15789,
        3581,
        30312,
        10115,
        235,
        10960,
        39810,
        45458,
        7626,
        11,
        10673,
        237,
        13547,
        17047,
        9890,
        98,
        2347,
        227,
        25243,
        7626,
        49161,
        9575,
        4511,
        17238,
        250,
        15433,
        8945,
        11,
        51315
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 34,
      "seek": 12004,
      "start": 139.04000000000002,
      "end": 143.72,
      "text": "两次提问都发生在播放过程中,后面的画面也没有被第一轮回答挡住。",
      "tokens": [
        51315,
        36257,
        9487,
        20949,
        22064,
        7182,
        28926,
        8244,
        3581,
        49993,
        12744,
        16866,
        29649,
        5975,
        11,
        13547,
        8833,
        1546,
        27126,
        8833,
        6404,
        17944,
        23238,
        18049,
        17819,
        106,
        8350,
        28223,
        8501,
        94,
        21632,
        1543,
        51549
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 35,
      "seek": 12004,
      "start": 143.72,
      "end": 147.16,
      "text": "这次运行也给出了一个很直接的硬件结论,",
      "tokens": [
        51549,
        5562,
        9487,
        3316,
        238,
        8082,
        6404,
        23197,
        7781,
        2289,
        20182,
        4563,
        43297,
        1546,
        38114,
        105,
        20485,
        45641,
        7422,
        118,
        11,
        51721
      ],
      "temperature": 0,
      "avg_logprob": -0.10400979895340769,
      "compression_ratio": 1.378917378917379,
      "no_speech_prob": 1.814225505936129e-11,
      "source": "whisper"
    },
    {
      "id": 36,
      "seek": 14716,
      "start": 147.16,
      "end": 152.28,
      "text": "我用两张4070Ti Super跑完了这段视频,扣掉系统本来的占用,",
      "tokens": [
        50365,
        1654,
        9254,
        36257,
        44059,
        5254,
        5867,
        51,
        72,
        4548,
        32585,
        41665,
        5562,
        28427,
        40656,
        39752,
        11,
        3416,
        96,
        29327,
        25368,
        10115,
        253,
        8802,
        6912,
        1546,
        5322,
        254,
        9254,
        11,
        50621
      ],
      "temperature": 0,
      "avg_logprob": -0.08952753360454853,
      "compression_ratio": 1.147239263803681,
      "no_speech_prob": 1.6501086955167388e-11,
      "source": "whisper"
    },
    {
      "id": 37,
      "seek": 14716,
      "start": 152.28,
      "end": 157.84,
      "text": "模型额外使用的显存大约13.3G,这个结果只对应当前样片和参数,",
      "tokens": [
        50621,
        41908,
        39823,
        12501,
        251,
        12022,
        22982,
        9254,
        1546,
        1431,
        122,
        39781,
        3582,
        16853,
        99,
        7668,
        13,
        18,
        38,
        11,
        15368,
        45641,
        9319,
        14003,
        8713,
        44297,
        16233,
        8945,
        14496,
        16668,
        12565,
        2129,
        224,
        33188,
        11,
        50899
      ],
      "temperature": 0,
      "avg_logprob": -0.08952753360454853,
      "compression_ratio": 1.147239263803681,
      "no_speech_prob": 1.6501086955167388e-11,
      "source": "whisper"
    },
    {
      "id": 38,
      "seek": 14716,
      "start": 157.84,
      "end": 162.04,
      "text": "但至少证明这套双卡配置可以完成一次本地实施推理。",
      "tokens": [
        50899,
        8395,
        20844,
        15686,
        5233,
        223,
        11100,
        5562,
        1881,
        245,
        2129,
        234,
        32681,
        38846,
        34719,
        6723,
        41509,
        27505,
        8802,
        10928,
        24726,
        4307,
        121,
        33597,
        13876,
        1543,
        51109
      ],
      "temperature": 0,
      "avg_logprob": -0.08952753360454853,
      "compression_ratio": 1.147239263803681,
      "no_speech_prob": 1.6501086955167388e-11,
      "source": "whisper"
    },
    {
      "id": 39,
      "seek": 14716,
      "start": 162.04,
      "end": 166.0,
      "text": "输出中间还有一段很值得看画面没有明显变化时,",
      "tokens": [
        51109,
        9830,
        241,
        7781,
        5975,
        31685,
        14852,
        32241,
        28427,
        4563,
        40242,
        5916,
        4200,
        27126,
        8833,
        17944,
        11100,
        1431,
        122,
        2129,
        246,
        23756,
        15729,
        11,
        51307
      ],
      "temperature": 0,
      "avg_logprob": -0.08952753360454853,
      "compression_ratio": 1.147239263803681,
      "no_speech_prob": 1.6501086955167388e-11,
      "source": "whisper"
    },
    {
      "id": 40,
      "seek": 14716,
      "start": 166.0,
      "end": 170.44,
      "text": "终端没有继续增加描述,后面的人物进入以后文字才重新出现。",
      "tokens": [
        51307,
        10115,
        230,
        11957,
        107,
        17944,
        10115,
        100,
        10115,
        255,
        24228,
        252,
        9990,
        11673,
        237,
        3316,
        108,
        11,
        13547,
        8833,
        29979,
        23516,
        36700,
        14028,
        3588,
        13547,
        17174,
        22381,
        18888,
        12624,
        12560,
        7781,
        20204,
        1543,
        51529
      ],
      "temperature": 0,
      "avg_logprob": -0.08952753360454853,
      "compression_ratio": 1.147239263803681,
      "no_speech_prob": 1.6501086955167388e-11,
      "source": "whisper"
    },
    {
      "id": 41,
      "seek": 17044,
      "start": 170.44,
      "end": 175.56,
      "text": "原始接口会把这种状态写成控制标记,真正做成应用时需要先过滤,",
      "tokens": [
        50365,
        19683,
        15476,
        14468,
        18144,
        12949,
        16075,
        5562,
        39810,
        35276,
        114,
        3757,
        223,
        5676,
        247,
        11336,
        48707,
        25491,
        162,
        3921,
        34756,
        11,
        6303,
        15789,
        10907,
        11336,
        44297,
        9254,
        15729,
        35748,
        10108,
        16866,
        15868,
        97,
        11,
        50621
      ],
      "temperature": 0,
      "avg_logprob": -0.11605156929262223,
      "compression_ratio": 1.2960526315789473,
      "no_speech_prob": 1.712588404423343e-11,
      "source": "whisper"
    },
    {
      "id": 42,
      "seek": 17044,
      "start": 175.56,
      "end": 178.76,
      "text": "对观众来说更重要的是模型知道什么时候先观察,",
      "tokens": [
        50621,
        8713,
        42350,
        7384,
        245,
        6912,
        8090,
        19002,
        24928,
        24620,
        41908,
        39823,
        7758,
        10440,
        29111,
        10108,
        42350,
        47550,
        11,
        50781
      ],
      "temperature": 0,
      "avg_logprob": -0.11605156929262223,
      "compression_ratio": 1.2960526315789473,
      "no_speech_prob": 1.712588404423343e-11,
      "source": "whisper"
    },
    {
      "id": 43,
      "seek": 17044,
      "start": 178.76,
      "end": 180.68,
      "text": "而不是每一帧都重复说话。",
      "tokens": [
        50781,
        11070,
        7296,
        23664,
        2257,
        4845,
        100,
        7182,
        12624,
        1787,
        235,
        8090,
        21596,
        1543,
        50877
      ],
      "temperature": 0,
      "avg_logprob": -0.11605156929262223,
      "compression_ratio": 1.2960526315789473,
      "no_speech_prob": 1.712588404423343e-11,
      "source": "whisper"
    },
    {
      "id": 44,
      "seek": 17044,
      "start": 180.68,
      "end": 184.68,
      "text": "不过,实时不代表每个瞬间都不会漏掉,当前测试每秒只取一帧,",
      "tokens": [
        50877,
        1960,
        16866,
        11,
        24726,
        15729,
        1960,
        19105,
        17571,
        23664,
        7549,
        36733,
        105,
        31685,
        7182,
        47928,
        14065,
        237,
        29327,
        11,
        16233,
        8945,
        11038,
        233,
        5233,
        243,
        23664,
        20956,
        14003,
        29436,
        2257,
        4845,
        100,
        11,
        51077
      ],
      "temperature": 0,
      "avg_logprob": -0.11605156929262223,
      "compression_ratio": 1.2960526315789473,
      "no_speech_prob": 1.712588404423343e-11,
      "source": "whisper"
    },
    {
      "id": 45,
      "seek": 17044,
      "start": 184.68,
      "end": 187.56,
      "text": "视频里一个非常短的击打动作就没有被稳定识别。",
      "tokens": [
        51077,
        40656,
        39752,
        15759,
        20182,
        14392,
        5881,
        255,
        1546,
        6336,
        119,
        12467,
        34961,
        11914,
        3111,
        17944,
        23238,
        10415,
        111,
        12088,
        5233,
        228,
        18453,
        1543,
        51221
      ],
      "temperature": 0,
      "avg_logprob": -0.11605156929262223,
      "compression_ratio": 1.2960526315789473,
      "no_speech_prob": 1.712588404423343e-11,
      "source": "whisper"
    },
    {
      "id": 46,
      "seek": 17044,
      "start": 187.56,
      "end": 191.56,
      "text": "它看到了动作前后不等于看见了中间的每一刻,",
      "tokens": [
        51221,
        11284,
        18032,
        2289,
        34961,
        11914,
        8945,
        13547,
        1960,
        10187,
        37732,
        4200,
        23813,
        2289,
        5975,
        31685,
        1546,
        23664,
        2257,
        45500,
        11,
        51421
      ],
      "temperature": 0,
      "avg_logprob": -0.11605156929262223,
      "compression_ratio": 1.2960526315789473,
      "no_speech_prob": 1.712588404423343e-11,
      "source": "whisper"
    },
    {
      "id": 47,
      "seek": 19156,
      "start": 191.56,
      "end": 196.68,
      "text": "提高采样频率可能保留更多细节,也会同时增加输入和缓存压力。",
      "tokens": [
        50365,
        20949,
        12979,
        5873,
        229,
        14496,
        39752,
        44866,
        16657,
        24302,
        24456,
        19002,
        6392,
        10115,
        228,
        45161,
        11,
        6404,
        12949,
        13089,
        15729,
        24228,
        252,
        9990,
        9830,
        241,
        14028,
        12565,
        38109,
        241,
        39781,
        5014,
        233,
        13486,
        1543,
        50621
      ],
      "temperature": 0,
      "avg_logprob": -0.06883793366261018,
      "compression_ratio": 1.2944297082228118,
      "no_speech_prob": 1.6910110464674055e-11,
      "source": "whisper"
    },
    {
      "id": 48,
      "seek": 19156,
      "start": 196.68,
      "end": 202.76,
      "text": "这种持续更新来自模型的输入方式,视觉信息不是等到整段结束再交给语言部分,",
      "tokens": [
        50621,
        5562,
        39810,
        17694,
        10115,
        255,
        19002,
        12560,
        6912,
        9722,
        41908,
        39823,
        1546,
        9830,
        241,
        14028,
        9249,
        27584,
        11,
        40656,
        24447,
        17665,
        26460,
        7296,
        10187,
        4511,
        27662,
        28427,
        45641,
        44320,
        8623,
        28455,
        23197,
        5233,
        255,
        12009,
        32174,
        11,
        50925
      ],
      "temperature": 0,
      "avg_logprob": -0.06883793366261018,
      "compression_ratio": 1.2944297082228118,
      "no_speech_prob": 1.6910110464674055e-11,
      "source": "whisper"
    },
    {
      "id": 49,
      "seek": 19156,
      "start": 202.76,
      "end": 208.68,
      "text": "每个采样画面还带着自己的时间位置,所以事件先后和状态变化能够进入同一段推理。",
      "tokens": [
        50925,
        23664,
        7549,
        5873,
        229,
        14496,
        27126,
        8833,
        14852,
        4845,
        99,
        20708,
        17645,
        1546,
        44848,
        11160,
        34719,
        11,
        7239,
        6973,
        20485,
        10108,
        13547,
        12565,
        35276,
        114,
        3757,
        223,
        2129,
        246,
        23756,
        8225,
        1787,
        253,
        36700,
        14028,
        13089,
        2257,
        28427,
        33597,
        13876,
        1543,
        51221
      ],
      "temperature": 0,
      "avg_logprob": -0.06883793366261018,
      "compression_ratio": 1.2944297082228118,
      "no_speech_prob": 1.6910110464674055e-11,
      "source": "whisper"
    },
    {
      "id": 50,
      "seek": 19156,
      "start": 208.68,
      "end": 213.56,
      "text": "官方代码除了读取本地视频还支持摄像头、屏幕和外部发送的视频帧,",
      "tokens": [
        51221,
        31929,
        9249,
        19105,
        23230,
        223,
        32999,
        2289,
        5233,
        119,
        29436,
        8802,
        10928,
        40656,
        39752,
        14852,
        35488,
        34783,
        226,
        12760,
        39862,
        1231,
        9636,
        237,
        40211,
        12565,
        12022,
        13470,
        28926,
        29309,
        1546,
        40656,
        39752,
        4845,
        100,
        11,
        51465
      ],
      "temperature": 0,
      "avg_logprob": -0.06883793366261018,
      "compression_ratio": 1.2944297082228118,
      "no_speech_prob": 1.6910110464674055e-11,
      "source": "whisper"
    },
    {
      "id": 51,
      "seek": 19156,
      "start": 213.56,
      "end": 219.0,
      "text": "它可以成为直播辅助实验观察、流程提醒或者屏幕任务监控的视觉底座。",
      "tokens": [
        51465,
        11284,
        6723,
        11336,
        13992,
        16186,
        49993,
        9830,
        227,
        37618,
        24726,
        49657,
        234,
        42350,
        47550,
        1231,
        27854,
        29649,
        20949,
        40796,
        31148,
        9636,
        237,
        40211,
        26443,
        5087,
        94,
        5419,
        239,
        48707,
        1546,
        40656,
        24447,
        22816,
        6346,
        100,
        1543,
        51737
      ],
      "temperature": 0,
      "avg_logprob": -0.06883793366261018,
      "compression_ratio": 1.2944297082228118,
      "no_speech_prob": 1.6910110464674055e-11,
      "source": "whisper"
    },
    {
      "id": 52,
      "seek": 21900,
      "start": 219.0,
      "end": 225.96,
      "text": "模型负责持续理解和回答,真正上线是告紧规则、权限控制和人工复合仍然要由应用层补上。",
      "tokens": [
        50365,
        41908,
        39823,
        18464,
        253,
        18464,
        96,
        17694,
        10115,
        255,
        13876,
        17278,
        12565,
        8350,
        28223,
        11,
        6303,
        15789,
        5708,
        16853,
        123,
        1541,
        16846,
        7732,
        100,
        6758,
        226,
        2437,
        247,
        1231,
        4422,
        225,
        43446,
        48707,
        25491,
        12565,
        4035,
        23323,
        1787,
        235,
        14245,
        1550,
        235,
        5823,
        4275,
        23786,
        44297,
        9254,
        9636,
        224,
        9890,
        98,
        5708,
        1543,
        50713
      ],
      "temperature": 0,
      "avg_logprob": -0.09523574120359994,
      "compression_ratio": 1.1909814323607426,
      "no_speech_prob": 1.4807471221400803e-11,
      "source": "whisper"
    },
    {
      "id": 53,
      "seek": 21900,
      "start": 225.96,
      "end": 232.04,
      "text": "所以Mousewell Real-Time的优势很明确,视频还在发声,问题和答案就能一起向前推进。",
      "tokens": [
        50713,
        7239,
        44,
        1316,
        6326,
        8467,
        12,
        22233,
        1546,
        7384,
        246,
        5087,
        123,
        4563,
        11100,
        38114,
        106,
        11,
        40656,
        39752,
        14852,
        3581,
        28926,
        32045,
        11,
        34069,
        12565,
        28223,
        28899,
        3111,
        8225,
        29567,
        24282,
        8945,
        33597,
        36700,
        1543,
        51017
      ],
      "temperature": 0,
      "avg_logprob": -0.09523574120359994,
      "compression_ratio": 1.1909814323607426,
      "no_speech_prob": 1.4807471221400803e-11,
      "source": "whisper"
    },
    {
      "id": 54,
      "seek": 21900,
      "start": 232.04,
      "end": 240.04,
      "text": "它的限制也同样明确,采样会漏掉短动作,长上下文会继续消耗缓存,原始输出也不是装好就能使用的监控产品。",
      "tokens": [
        51017,
        45224,
        43446,
        25491,
        6404,
        13089,
        14496,
        11100,
        38114,
        106,
        11,
        5873,
        229,
        14496,
        12949,
        14065,
        237,
        29327,
        5881,
        255,
        34961,
        11914,
        11,
        32271,
        5708,
        4438,
        17174,
        12949,
        10115,
        100,
        10115,
        255,
        28837,
        4450,
        245,
        38109,
        241,
        39781,
        11,
        19683,
        15476,
        9830,
        241,
        7781,
        6404,
        7296,
        8083,
        227,
        2131,
        3111,
        8225,
        22982,
        9254,
        1546,
        5419,
        239,
        48707,
        1369,
        100,
        30246,
        1543,
        51417
      ],
      "temperature": 0,
      "avg_logprob": -0.09523574120359994,
      "compression_ratio": 1.1909814323607426,
      "no_speech_prob": 1.4807471221400803e-11,
      "source": "whisper"
    },
    {
      "id": 55,
      "seek": 21900,
      "start": 240.04,
      "end": 245.48,
      "text": "如果你的任务只是等视频结束以后再总结,指令版会更直接。",
      "tokens": [
        51417,
        13119,
        18961,
        26443,
        5087,
        94,
        36859,
        10187,
        40656,
        39752,
        45641,
        44320,
        3588,
        13547,
        8623,
        33440,
        45641,
        11,
        25922,
        49061,
        42096,
        12949,
        19002,
        43297,
        1543,
        51689
      ],
      "temperature": 0,
      "avg_logprob": -0.09523574120359994,
      "compression_ratio": 1.1909814323607426,
      "no_speech_prob": 1.4807471221400803e-11,
      "source": "whisper"
    },
    {
      "id": 56,
      "seek": 24548,
      "start": 245.48,
      "end": 252.44,
      "text": "如果你需要在视频进行中观察、提问等待,再根据新画面更新判断,实施版才真正对口。",
      "tokens": [
        50365,
        45669,
        35748,
        3581,
        40656,
        39752,
        36700,
        8082,
        5975,
        42350,
        47550,
        1231,
        20949,
        22064,
        10187,
        18390,
        11,
        8623,
        31337,
        26075,
        106,
        12560,
        27126,
        8833,
        19002,
        12560,
        2437,
        97,
        4307,
        255,
        11,
        24726,
        4307,
        121,
        42096,
        18888,
        6303,
        15789,
        8713,
        18144,
        1543,
        50713
      ],
      "temperature": 0,
      "avg_logprob": -0.10718824646689674,
      "compression_ratio": 0.9416666666666667,
      "no_speech_prob": 1.662408405378457e-11,
      "source": "whisper"
    },
    {
      "id": 57,
      "seek": 25244,
      "start": 252.44,
      "end": 257.4,
      "text": "项目地址和官方资料放在评论区,感兴趣可以拿自己的视频试一遍。",
      "tokens": [
        50365,
        10178,
        117,
        11386,
        10928,
        14872,
        222,
        12565,
        31929,
        9249,
        5266,
        226,
        33404,
        12744,
        3581,
        5233,
        226,
        7422,
        118,
        9937,
        118,
        11,
        9709,
        2347,
        112,
        39835,
        6723,
        24351,
        17645,
        1546,
        40656,
        39752,
        5233,
        243,
        2257,
        3330,
        235,
        1543,
        50613
      ],
      "temperature": 0,
      "avg_logprob": -0.10118260383605956,
      "compression_ratio": 0.8888888888888888,
      "no_speech_prob": 1.526043180710701e-11,
      "source": "whisper"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "zh",
    "duration": 258.88,
    "reference_path": null,
    "reference_kind": null,
    "reference_language": null,
    "selected_source": "whisper_segmented",
    "decision_reason": "source_unavailable_or_untrusted_whisper_passed",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 0,
      "duration_coverage": 0.0,
      "usable_full": false,
      "usable_partial": false,
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "language": {
        "requested": "zh",
        "matched": false
      }
    },
    "whisper_profile": {
      "segment_count": 58,
      "multilingual": false,
      "first_start": 0.0,
      "final_end": 257.4,
      "duration_coverage": 0.994283,
      "maximum_gap_seconds": 0.0,
      "inverted_count": 0,
      "language": {
        "requested": "zh",
        "matched": true,
        "latin": 51,
        "han": 1383,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "comparison": {
      "window_seconds": 30.0,
      "compared_window_count": 0,
      "median_score": 0.0,
      "divergent_window_count": 0,
      "divergent_ratio": 0.0,
      "divergent_windows": []
    },
    "timestamp_alignment": {
      "applied": false,
      "estimated_offset_seconds": 0.0,
      "matched_segment_count": 0,
      "match_scores": []
    },
    "caption_timing_repair": null,
    "filled_source_profile": null,
    "whisper_duration_guard": {
      "media_duration": 258.88,
      "rejected_segment_count": 0,
      "rejected_segments": []
    }
  }
}