ElevenLabsSpeechToText - ComfyUI Built-in Node Documentation - ComfyUI

The ElevenLabs Speech to Text node transcribes audio into text using the ElevenLabs API. It supports automatic language detection, speaker diarization (identifying different speakers), and audio event tagging (annotating sounds like laughter or music in the transcript).

Inputs

Common Inputs

Parameter	Description	Data Type	Required	Range
`model`	Model to use for transcription. Selecting this model reveals additional parameters.	DYNAMIC_COMBO	Yes	`"scribe_v2"`
`audio`	Audio to transcribe.	AUDIO	Yes	-
`language_code`	ISO-639-1 or ISO-639-3 language code (e.g., ‘en’, ‘es’, ‘fra’). Leave empty for automatic detection. (default: "")	STRING	No	-
`num_speakers`	Maximum number of speakers to predict. Set to 0 for automatic detection. (default: 0)	INT	No	0 - 32
`seed`	Seed for reproducibility (determinism not guaranteed). (default: 1)	INT	No	0 - 2147483647

Scribe v2 Inputs

These parameters are shown when the "scribe_v2" model is selected.

Parameter	Description	Data Type	Required	Range
`tag_audio_events`	Annotate sounds like (laughter), (music), etc. in transcript. (default: False)	BOOLEAN	No	-
`diarize`	Annotate which speaker is talking. (default: False)	BOOLEAN	No	-
`diarization_threshold`	Speaker separation sensitivity. Lower values are more sensitive to speaker changes. Only used when `diarize` is enabled. (default: 0.22)	FLOAT	No	0.1 - 0.4
`temperature`	Randomness control. 0.0 uses model default. Higher values increase randomness. (default: 0.0)	FLOAT	No	0.0 - 2.0
`timestamps_granularity`	Timing precision for transcript words. (default: “word”)	COMBO	No	`"word"` `"character"` `"none"`

Note: num_speakers cannot be set to a value greater than 0 when diarize is enabled. You must either disable diarize or set num_speakers to 0.

Outputs

Output Name	Description	Data Type
`text`	The transcribed text from the audio.	STRING
`language_code`	The detected language code of the audio.	STRING
`words_json`	A JSON-formatted string containing detailed word-level information, including timestamps and speaker labels if enabled.	STRING

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 7eb5d72615aa8a9e4a8014e45b39cf83dc8d8432d7ce0dccba20489be80a5830