#5594 (18.666667 > 2.400000)2026-07-02 16:19:45,208 - DEBUG - Compression ratio threshold is not met with temperature 0.6 (

114.233* Posted at: 7 hours ago

(18.666667 > 2.400000)
2026-07-02 16:19:45,208 - DEBUG - Compression ratio threshold is not met with temperature 0.6 (18.666667 > 2.400000)
2026-07-02 16:19:49,907 - DEBUG - Compression ratio threshold is not met with temperature 0.8 (18.666667 > 2.400000)
2026-07-02 16:19:54,608 - DEBUG - Compression ratio threshold is not met with temperature 1.0 (18.666667 > 2.400000)
2026-07-02 16:19:54,958 - DEBUG - Processing segment at 05:00.000
2026-07-02 16:20:03,948 - DEBUG - Compression ratio threshold is not met with temperature 0.0 (18.666667 > 2.400000)
2026-07-02 16:20:03,948 - DEBUG - Log probability threshold is not met with temperature 0.0 (-291.128876 < -1.000000)
2026-07-02 16:20:08,682 - DEBUG - Compression ratio threshold is not met with temperature 0.2 (18.666667 > 2.400000)
2026-07-02 16:20:13,535 - DEBUG - Compression ratio threshold is not met with temperature 0.4 (18.666667 > 2.400000)
2026-07-02 16:20:18,574 - DEBUG - Compression ratio threshold is not met with temperature 0.6 (18.666667 > 2.400000)
2026-07-02 16:20:23,382 - DEBUG - Compression ratio threshold is not met with temperature 0.8 (18.666667 > 2.400000)
2026-07-02 16:20:28,208 - DEBUG - Compression ratio threshold is not met with temperature 1.0 (18.666667 > 2.400000)
2026-07-02 16:20:28,608 - DEBUG - Processing segment at 05:30.000

1 Replies

Floor #1 DeepSeek reply at 7 hours ago

这个日志是 Whisper 语音识别模型在尝试转录时，因为音频质量或背景噪声问题导致多次温度采样均不满足压缩比或对数概率阈值，从而无法输出有效文本。模型反复尝试（0.0~1.0）仍未通过，常见于以下情形：

  1. 音频片段中无有效人声（如纯背景音乐、噪音、静音或极低音量）
  2. 发音语言选择错误（例如视频是日语但你选了中文）
  3. 环境噪声过大，VAD 未正确过滤

解决方案

  1. 确认发音语言：在主界面正确选择 Spoken Lang（发音语言），不要使用 auto。
  2. 启用降噪：点击 设置更多参数 → 勾选 降噪，或勾选 分离人声背景声 以获得更干净的语音。
  3. 调整 VAD 参数（可选）：进入 菜单 → 工具 → 高级选项 → 语音识别参数，将 语音阈值 适当调低（如从 0.5 降至 0.3），非语音阈值 适当调高。
  4. 更换识别渠道或模型：中文推荐 Qwen-ASR(本地内置) 或 字节语音识别大模型极速版，英文推荐 faster-whisper(本地内置) large-v3。
  5. 检查源文件：用播放器确认该时间段确有清晰人声，若无则忽略该段。

请查阅相关文档：

Post Your Reply
Open source and free maintenance is not easy. If this project is helpful to you, please consider making a small donation to help the project continue to maintain and update. Dondate: ko-fi.com/jianchang512

Related resource