Speech-to-Text (Audio Transcription)

Powered by the Whisper large-v3 model via the LemonFox API. Converts audio files or public audio URLs into text transcripts with support for 100+ languages, speaker recognition, word-level timestamps, and multiple output formats.

Billing: Billed per API call.

---

Overview

Property	Value
Endpoint	POST https://app-dohnbj5by5mp-api-DY8MNQoqOnMa.gateway.appmedo.com/v1/audio/transcriptions
Auth	Bearer Token (platform_managed, X-Gateway-Authorization header)
Content-Type	multipart/form-data (file upload) or application/x-www-form-urlencoded (URL submission)
Max File Size	Upload: 100 MB; URL submission: 1 GB
Supported Formats	mp3, wav, flac, aac, opus, ogg, m4a, mp4, mpeg, mov, webm, and more
Supported Languages	100+, auto-detected if not specified
Billing	Billed per call

Core capabilities:

• Audio-to-text transcription, returning plain text or detailed JSON with timestamps
• Multi-speaker recognition (Speaker Diarization), automatically labels each speaker
• Multiple output formats: json, text, srt, vtt, verbose_json
• Supports async callback via callback_url
• Supports translating audio content to English

For complete parameter details, response fields, and code examples, see references/transcriptions-api.md.

---

Generation-time Usage (Agent Direct Call)

Use the built-in script for generation-time calls. The script reads INTEGRATIONS_API_KEY from the environment.

The Bash tool timeout MUST be set to 600000ms (600 seconds).

# From a public audio URL (recommended, supports up to 1 GB)
python3 <skill-path>/scripts/call_speech_to_text.py \
  --file-url "https://example.com/audio.mp3" \
  --response-format json

# From a local audio file (multipart upload, up to 100 MB)
python3 <skill-path>/scripts/call_speech_to_text.py \
  --file /path/to/audio.mp3 \
  --response-format verbose_json \
  --speaker-labels --word-timestamps

# Translate audio to English
python3 <skill-path>/scripts/call_speech_to_text.py \
  --file-url "https://example.com/chinese_audio.wav" \
  --translate

The script prints one JSON line on success: {"status":"succeed","result":{...}} (where result is the upstream response). For text, srt, or vtt formats, the result is {"text":"..."}. On failure it prints an error to stderr and exits with a non-zero code.

For complete parameter details and response field tables, see references/transcriptions-api.md.

---

Post-generation Usage (In-app via Edge Function)

When called from within an application, requests must be proxied through an Edge Function to keep the platform key out of the frontend.

For the complete Edge Function code and frontend call patterns, see the "Post-generation Usage" section in references/transcriptions-api.md.