ComfyUI Node
(Deno) Audio Transcript
DENO Custom Nodes v0.7.88. The DENO info button checks Comfy Registry and marks this node when an update is available. Transcribes one source audio clip locally with official Whisper, using CUDA Smart Swap and first-use model download, then returns structured context plus the effective transcript. Optional user-supplied lyrics or dialogue overrides Whisper wording while Whisper still supplies approximate segment timing.
(Deno) Audio Transcript
- audio
- audio_context
- transcript
- audio
◄modellarge-v3-turbo►
◄languageauto►
◄model_after_runUnload after run►
◄manual_transcript►
CategoryDeno/Audio
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| audio | AUDIO | Source mono or stereo audio to transcribe locally. The node downmixes it to mono and resamples it to 16 kHz for Whisper. | |
| model | COMBO | large-v3-turbo | Local Whisper model size. large-v3 prioritizes lyric and difficult-speech accuracy; large-v3-turbo is the faster default. The official checkpoint downloads on first use to ComfyUI/models/stt/whisper. |
| language | COMBO | auto | Choose a known language or let Whisper detect it automatically. |
| model_after_run | COMBO | Unload after run | CUDA Smart Swap unloads ComfyUI models before transcription. Unload also releases Whisper afterward; Keep loaded is an advanced repeated-run option. |
| manual_transcriptopt | STRING | Optional exact lyrics or dialogue. When non-empty, this wording is authoritative while Whisper still runs to provide approximate segment timing. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| audio_context | STRING | Structured transcript data with requested/detected language, timestamped segments, and a heuristic confidence band for an LLM prompt builder. |
| transcript | STRING | Effective plain transcript: the exact user-supplied lyrics or dialogue when provided, otherwise the Whisper transcript. |
| audio | AUDIO | The original source AUDIO unchanged, for a guaranteed Whisper-before-Gemma Smart Swap chain. |