Google Vertex AI icon

Google Vertex AI

Use Google Vertex AI (Gemini) for text, media analysis, and image generation/editing

Actions11

Audio → Transcribe Audio

AI-generated

Summary

Transcribes an audio file to text by sending it, with a prompt, to a Gemini model on Google Vertex AI (generateContent). Audio comes from an incoming binary property or a Google Cloud Storage URI.

Inputs

  • Model (required) — Gemini model, picked from a list (e.g. gemini-2.5-flash, the default) or entered by ID.
  • Input Type — Binary File (default) or Google Cloud Storage URI.
  • Input Binary Field — Binary property name holding the audio (default 'data'). Used when Input Type is Binary File.
  • GCS URI — gs://bucket/path/file.ext. Used when Input Type is Google Cloud Storage URI.
  • MIME Type — Optional for binary, where it overrides the detected type. Required for a GCS URI, or the node errors.
  • Prompt (required) — Instruction for the model. Defaults to 'Generate a complete, accurate transcript of the speech in this audio.'
  • Options — Optional settings: JSON Output (requests application/json), Max Output Tokens (default 1024), Simplify Output (default true), System Instruction, Temperature, Top K and Top P.

Output shape

One output item per input item. With Simplify Output on (the default), the item is {text: }. With it off, the item is the full Vertex AI generateContent response.

Max Output Tokens defaults to 1024, so long recordings may be cut off unless you raise it. The transcript comes from the model's prompt response, not a dedicated speech-to-text API. There are no timestamps or speaker labels unless you ask for them in the prompt. If Continue On Fail is enabled, a failed item returns {error: message}. Authentication uses the Google Vertex AI service-account credential.

Examples

Example 1: Transcribe a voice memo received from a webhook or file download node.

Input Type: Binary File; Input Binary Field: data; default Prompt; Model: gemini-2.5-flash; Max Output Tokens raised to 8192.

Example 2: Transcribe a recording stored in Cloud Storage and return structured JSON.

Input Type: Google Cloud Storage URI; GCS URI: gs://my-bucket/call.mp3; MIME Type: audio/mp3; Prompt asks for speaker-labeled segments; Options: JSON Output on.

Links

Discussion