Nodes/CRT-Nodes/Audio Transcript Batch (CRT)
ComfyUI Node

Audio Transcript Batch (CRT)

Transcribes a batched AUDIO input with faster-whisper (CTranslate2). Outputs one transcript per batch item as a list that pairs item by item with the batch loader's file_names in SaveTextWithPath.

By PGCRT·Created 2 years ago·Updated about 11 hours ago· 132
Audio Transcript Batch (CRT)
  • audio
  • strings_batch
  • display
  • status
isolate_voicefalse
keep_model_loadedtrue
whisper_modellarge-v3-turbo
compute_typefloat16
vad_filtertrue
transcribe_batch_size8
beam_size5
languageauto
CategoryCRT/Audio

Inputs (9)

NameTypeDefaultDescription
audioAUDIO
isolate_voiceBOOLEANfalseRun MelBandRoFormer voice isolation on each file before transcription.
keep_model_loadedBOOLEANtrueKeep the Whisper model loaded in VRAM after batch transcription. Disable to free VRAM immediately.
whisper_modeloptCOMBOlarge-v3-turbo14 options: tiny.en, tiny, base.en, base, small.en, small, +8
compute_typeoptCOMBOfloat164 options: float16, int8_float16, int8, float32
vad_filteroptBOOLEANtrueCuts silences before inference for faster long-audio processing. Uses Silero VAD to skip silent regions.
transcribe_batch_sizeoptINT81–64Number of 30s chunks decoded in parallel per file. 1 disables batched decoding.
beam_sizeoptINT51–10Beam search width. 1 = greedy decoding (faster, slightly less accurate).
languageoptSTRINGauto

Outputs (3)

NameTypeDescription
strings_batchSTRINGOne transcript per batch item as a list — pairs item by item with the batch loader's file_names in SaveTextWithPath.
displaySTRINGSame transcripts stitched into one display string, separated by a blank line per item.
statusSTRINGSummary status for the batch run.