ComfyUI SupaVoice Transcriber

Adds transcript/labeling nodes under SupaVoice/audio.

Nodes:

• SupaVoice Media Audio Trim
  • Extracts audio from an audio/video file path using start_seconds and duration_seconds.
  • No preview player or autoplay UI.
  • Returns ComfyUI AUDIO plus the saved trimmed WAV path.
• SupaVoice Video Loader
  • Loads video frames with ffmpeg without browser preview audio.
  • Supports start_seconds, end_seconds, duration_seconds, fps, width, height, and max_frames.
  • Outputs ComfyUI IMAGE frames, clean AUDIO, saved audio path, and metadata JSON.
• SupaVoice Audio to WAV
  • Converts ComfyUI/VHS audio into a plain ComfyUI AUDIO dict and saved WAV file.
  • Useful as an adapter before nodes that reject Video Helper Suite LazyAudioMap.
  • Can force mono/stereo and resample.
• SupaVoice WAV/Media to Audio
  • Loads audio from a WAV/audio/video path into ComfyUI AUDIO.
  • Can also trim by start/duration while loading.
• SupaVoice Extract Vocals
  • Uses Demucs locally through the Python API.
  • Returns isolated vocals AUDIO plus saved vocals_path and other_path.
  • Use before transcription when source media has music/background audio.
  • Accepts normal ComfyUI AUDIO and Video Helper Suite lazy audio outputs.
  • Saves with soundfile to avoid the Windows TorchCodec/torchaudio save path.
• SupaVoice Local Whisper Transcriber
  • Uses faster-whisper.
  • Good for local/offline English and first-pass Burmese.
• SupaVoice HF Whisper Transcriber
  • Uses Hugging Face transformers ASR models directly.
  • Includes chuuhtetnaing Myanmar Whisper presets: tiny, small, medium, and large-v3.
• SupaVoice Gemini/OpenRouter Transcriber
  • Uses OpenRouter audio input with Gemini models.
  • Better for Burmese labeling when local Whisper is weak.
• SupaVoice Save Transcript
  • Saves edited transcript text to a .txt file.

Outputs:

• transcript: plain text to connect into VoxCPM prompt/training fields.
• txt_path: saved text transcript.
• json_path: saved detailed result.

Files are saved under:

E:\SupaVoiceData\comfy_transcripts
E:\SupaVoiceData\comfy_media_trim
E:\SupaVoiceData\comfy_audio_convert
E:\SupaVoiceData\comfy_voice_extract

For OpenRouter/Gemini, set the key in:

E:\git\SupaVoice\.env

Accepted key names: OPENROUTER_API_KEY, OPEN_ROUTER, or OPENROUTER.

Install location:

X:\git\ComfyUI\custom_nodes\ComfyUI-SupaVoice-Transcriber

Restart ComfyUI after adding or updating the node.

Notes:

• Local Whisper in the current comfy env needs a Python 3.12 compatibility shim for older setuptools; the node includes it.
• The chuuhtetnaing Myanmar Whisper cards report high WER overall, but small/medium are worth local testing and may beat generic Whisper on some Burmese clips.
• For difficult Burmese clips, Gemini/OpenRouter is expected to be much better than local small Whisper.
• Saved transcripts can be edited, then passed into VoxCPM prompt/training fields.