Speech to Text Tools

The Speech to Text provider transcribes audio or video files into text using the ElevenLabs speech-to-text API. Use it in a workflow when you have a publicly accessible audio/video URL and need a transcript (with word-level segments and detected language) to feed into downstream blocks such as agents, summarizers, or storage.

Overview

Operations

Configuration

stt_transcribe

Calls POST https://api.elevenlabs.io/v1/speech-to-text with the xi-api-key header for authentication. The request is sent as application/x-www-form-urlencoded, passing the audio URL as cloud_storage_url and requesting word-level timestamp granularity.

Outputs

stt_transcribe

• data (json) — Transcript text, segments, and detected language. Contains:
  • transcript (string) — The full transcribed text.
  • segments (array of json) — Word-level segments, each with text, start, end, and speaker.
  • language (string | null) — The detected or specified language code.
• metadata (json) — Transcription metadata. Contains:
  • language (string) — Detected or specified language code.
  • segmentCount (number) — Number of word-level segments returned.

YAML Example

stt_1:
  type: stt
  name: "Speech to Text"
  inputs:
    operation: "stt_transcribe"
    audioUrl: "https://example.com/audio.mp3"
    model: "scribe_v2"
    language: "auto"
    apiKey: "{{ELEVENLABS_API_KEY}}"
  connections:
    outgoing:
      - target: next-block-id

Tips

• Authenticate with an ElevenLabs API key — set it via an environment variable and reference it as {{ELEVENLABS_API_KEY}} rather than hardcoding the secret.
• The audioUrl must be a publicly accessible URL; ElevenLabs fetches it server-side as cloud_storage_url, so signed/expiring links may need to stay valid for the duration of the call.
• Leave language empty or set it to auto to let ElevenLabs auto-detect the spoken language; only pass a specific code (e.g. en) when you need to force it.
• Downstream blocks can consume the transcript with {{stt_1.data.transcript}}, the detected language with {{stt_1.data.language}}, and the segment count with {{stt_1.metadata.segmentCount}}.