WARNING:

This is currently not working as intended and is not developed further. Timings are not quite correct and the model get's stuck sometimes.

Vox Audio Transcription

Audio transcription using Mistral AI's Voxtral API. Supports long audio files and provides a REST API for job-based processing.

Features

• Transcribes audio/video files using Mistral AI
• Handles long files by splitting into segments
• REST API with job queue
• Outputs SRT subtitles or JSON format
• Docker support

Installation

Using UV

git clone <repository>
uv sync

Using Docker

docker build -t transcription-api .
docker run -d -p 8888:8888 -e MISTRAL_API_KEY=your_key transcription-api

Web API Usage

Start the server:

MISTRAL_API_KEY=your_key uv run python web_server.py

API Endpoints

Create Transcription Job

POST /transcription/job
Content-Type: application/json

{
  "path": "/path/to/audio.mp3",
  "language": "en",
  "format": "json",
  "priority": 1000
}

Response:

{
  "job_id": "uuid-string",
  "status": "QUEUED"
}

Get Job Status

GET /transcription/job/{job_id}

Response (completed):

{
  "job_id": "uuid-string",
  "status": "COMPLETED",
  "result": {
    "text": "Full transcription...",
    "segments": [
      {
        "text": "Segment text",
        "start": 0.0,
        "end": 5.2,
        "words": [...]
      }
    ]
  }
}

List All Jobs

GET /transcription/jobs

Get Queue Stats

GET /stats

Configuration

• Supported formats: json, srt
• Supported models: voxtral-small-2507, voxtral-small-latest
• Supported languages: All 67 Mistral AI language codes
• Priority: 1-499 (low), 500-9999 (high)

Requirements

• Python 3.9+
• Mistral AI API key
• FFmpeg (for audio processing)