Groq Speech-to-TextInstall

Summary

Translate audio input from a URL or binary data into English text using Groq Whisper models, returning the translation in customizable formats with optional transcription guidance and timestamp granularity.

Inputs

• Input Type (required) — Select whether the audio file is provided via a URL or as raw binary data in the workflow.
• Audio URL (required) — Provide the URL of the audio file to be translated when 'Audio URL' input type is selected. Supported formats include mp3, mp4, wav, m4a, flac, ogg, and webm.
• Binary Property (required) — Specify the binary property name containing the audio file when 'Binary Data' input type is selected; defaults to 'data'.
• Model (required) — Choose between two Whisper models: 'Whisper Large V3 Turbo' for faster, lower-cost processing or 'Whisper Large V3' for higher accuracy.
• Language — Optionally specify the audio's language using an ISO-639-1 code to improve translation accuracy; if omitted, language is auto-detected.
• Response Format (required) — Select the transcription response format: 'JSON' for simple text, 'Verbose JSON' for metadata and timestamps (default), or 'Text' for plain text only.
• Additional Options — Configure optional settings including a prompt to guide transcription (max 224 tokens), temperature (sampling randomness between 0 and 1), and timestamp granularities ('segment' and/or 'word') to include timestamps at different detail levels.

Output shape

A list of JSON objects each containing the translated text and metadata.

Each item in the output corresponds to one input audio item, including the translation (in the chosen format), metadata about the model, operation, language, input type, and original file name. On errors, output includes an error message object if 'Continue On Fail' is enabled; otherwise, execution stops with error. No pagination is involved.