ComfyUI Node
ThinkingLLM Whisper ASR
A ComfyUI node in ThinkingLLM/Audio with 9 inputs and 3 outputs.
ThinkingLLM Whisper ASR
- audio
- TRANSCRIPT
- SEGMENTS_JSON
- RAW_TRACE
◄model_sizesmall►
◄languageauto►
◄tasktranscribe►
◄devicecpu►
◄compute_typeint8►
◄beam_size5►
◄vad_filtertrue►
◄audio_file_path►
CategoryThinkingLLM/Audio
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| model_size | COMBO | small | faster-whisper model size. small is a reliable Windows-friendly starting point; large-v3 is higher quality but downloads and runs much larger. |
| language | COMBO | auto | Audio language. Use auto to let Whisper detect it. |
| task | COMBO | transcribe | transcribe preserves the original language; translate returns English. |
| device | COMBO | cpu | cpu is the most reliable Windows default; choose cuda when your faster-whisper/CTranslate2 CUDA runtime is working. |
| compute_type | COMBO | int8 | int8 is reliable on CPU. Use float16 or int8_float16 for CUDA. |
| beam_size | INT | 51–10 | Higher values can improve accuracy but run slower. |
| vad_filter | BOOLEAN | true | Skip long silence and non-speech sections before transcription. |
| audioopt | AUDIO | — | |
| audio_file_pathopt | STRING | Optional local audio file path. M4A, MP3, WAV, FLAC, and other FFmpeg-readable files are decoded to 16 kHz mono WAV before Whisper. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| TRANSCRIPT | STRING | — |
| SEGMENTS_JSON | STRING | — |
| RAW_TRACE | STRING | — |