Transcriptions

a full stack web app built with Svelte, Tailwind, Flask that can be self hosted with docker compose, and offers a simple way to transcribe audio files

Features

• Supports 4 model providers:
  • Groq (inferencing OpenAI Whisper) - cheapest, medium speed, medium accuracy
  • Google (through gemini audio multimodality) - cheapest/medium price, slower speed, high accuracy
  • FireworksAI (inferencing OpenAI Whisper) - medium price, fastest speed, semi-high accuracy
  • ElevenLabs (in-house Scribe model) - most expensive, slowest speed, highest accuracy
• Saves a transcription history
• Saves settings between frontend visits
• Supports multiple audio file formats
• Supports multiple languages
• Supports multiple speaker diarization
• Can return word level or segment level timestamps
• BYOK - Bring your own keys

Preview

Setup

Option 1: Docker Setup (Recommended)

Prerequisites

• Docker
• Docker Compose

Clone the repository

git clone https://github.com/nocdn/transcriptions.git
cd transcriptions

Configure your API keys and settings by copying the .env.example file to .env and filling it in with your own values.

cp .env.example .env

Build and start the Docker containers.

docker compose up -d --build

Depending on where you host it, the application will be available at http://localhost:4090 or http://[your-machine-ip]:4090

Updating (if using Docker)

Stop the containers:

docker compose down

Pull the latest changes:

git pull

Build and start the Docker containers again:

docker compose up -d --build

Option 2: Local Development

Prerequisites

You must have the following installed on your machine:

• npm
• Python

Clone the repository:

git clone https://github.com/nocdn/transcriptions.git

Install the frontend dependencies:

cd transcriptions/frontend
npm install

From the same directory, start the front-end development server:

npm run dev

For the backend:

Configure your API keys and settings by copying the .env.example file to .env and filling it in with your own values.

cd transcriptions/backend
cp .env.example .env

In a new terminal, start the back-end development server:

cd transcriptions/backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python app.py

Using the API

If you've exposed the backend service (either via Docker port mapping or running locally), you can interact with the API directly. The base URL will be http://<your-host-ip>:6005/api.

Endpoints

1. Upload & Transcribe Audio

   • URL: /upload
   • Method: POST
   • Content-Type: multipart/form-data
   • Form Data:
     • file: The audio file to transcribe (required).
     • currentModelProvider: The provider to use: groq, gemini, fireworks, or elevenlabs (required).
     • Provider Specific Fields (required based on currentModelProvider):
       • If groq: groqModelValue, groqPromptValue (optional), groqLanguageValue.
       • If gemini: geminiModelValue, geminiPromptValue (optional, defaults to value in .env).
       • If fireworks: fireworksModelValue, fireworksLanguageValue.
       • If elevenlabs: elevenlabsModelValue.
     • webhookUrl: A URL to send the transcription result to via a POST request (optional).
   • Success Response (200 OK):

     {
       "transcription": "The transcribed text..."
     }

   • Error Responses:
     • 400 Bad Request: Missing file, invalid provider, or unsupported file type.
     • 429 Too Many Requests: Rate limit exceeded.
     • 500 Internal Server Error: Backend processing error.
   • Example (using curl):

     curl -X POST -F "file=@/path/to/your/audio.mp3" \
          -F "currentModelProvider=groq" \
          -F "groqModelValue=whisper-large-v3" \
          -F "groqLanguageValue=en" \
          http://<your-host-ip>:6005/api/upload

2. Get Transcription History

   • URL: /history
   • Method: GET
   • Success Response (200 OK): An array of history items.

     [
       {
         "date": "26-07-2024",
         "fileExtension": "mp3",
         "fileNameNoExt": "myaudio",
         "filename": "myaudio.mp3",
         "transcription": "The transcribed text..."
       },
       // ... more items
     ]

   • Error Responses:
     • 500 Internal Server Error: Error reading history.
   • Example (using curl):

     curl http://<your-host-ip>:6005/api/history

3. Delete History Item

   • URL: /delete
   • Method: POST
   • Content-Type: application/json
   • Request Body:

     {
       "filename": "myaudio.mp3"
     }

     (Note: Use the original filename as returned by the /history endpoint)
   • Success Response (200 OK):

     {
       "success": true,
       "message": "Deleted myaudio.mp3"
     }

   • Error Responses:
     • 400 Bad Request: Missing filename in request body.
     • 404 Not Found: The specified file doesn't exist in the history.
     • 500 Internal Server Error: Error during deletion.
   • Example (using curl):

     curl -X POST -H "Content-Type: application/json" \
          -d '{"filename": "myaudio.mp3"}' \
          http://<your-host-ip>:6005/api/delete

Rate Limiting

The API endpoints have rate limits configured in the .env file (TRANSCRIPTION_RATE_LIMIT, HISTORY_RATE_LIMIT, DELETE_RATE_LIMIT). These limits are applied per IP address. If you exceed the limit, you will receive a 429 Too Many Requests response.

License

This project is licensed under the MIT License - see the LICENSE file for details