whisper-cpp 2

Voice input for the DeepSeek Harness Web Client: a microphone control in the composer, a transcription capability seam, and two selectable providers

Continuous video frame sampling plus local VAD-gated transcription for DeepSeek Harness, aligned on one timeline. · 为 DSH 提供连续视频抽帧与本地离线语音转写，画面与语音按时间轴对齐。