Transcribe
AI-generatedSummary
Transcribe speech from a WAV or MP3 audio file within the incoming binary data into English text using a specified Whisper model.
Inputs
- Audio Input Type (required) — Select 'Binary File' to provide the audio as a WAV or MP3 file within the incoming item's binary data.
- Binary Property Name (required) — Specify the name of the binary property in the incoming item containing the audio file (e.g., 'data'). If not found, the node attempts to use the only available binary property if exactly one exists.
- Model (required) — Choose the English speech recognition model to apply (e.g., Xenova/whisper-tiny.en, whisper-base.en, whisper-small.en, or whisper-medium.en). Smaller models are faster and use less memory.
Output shape
a list of items, each corresponding to an input item, with JSON data augmented by a 'transcription' field containing the recognized English text from the audio file
The node accepts WAV and MP3 audio formats only; it resamples audio to 16kHz mono as needed. If the binary property specified is missing and multiple properties exist, an error is thrown unless 'continue on fail' is enabled. Errors for individual items are returned in an error envelope with the 'error' property and paired item index. The transcription process chunk-length defaults to 30 seconds with a stride of 5 seconds.
Examples
Example 1: Transcribe audio from an incoming WAV file in the binary property 'data' using the Xenova/whisper-small.en model.
Set Audio Input Type to 'Binary File', Binary Property Name to 'data', and Model to 'Xenova/whisper-small.en'. Connect a node that supplies WAV or MP3 audio as binary data under property 'data' to this node.