Voice2Text Web Service

This project is a web service built using Flask, Python, Nginx, and Docker, aimed at converting voice input to text using trained models.

based on [1] by Valerio Velardo.

Project Structure

voice2text/
│
├── flask/
│   ├── paraphraser.h5/
│   ├── processor.h5/
│   ├── voice2text.h5/
│   ├── app.ini
│   ├── Dockerfile
│   ├── requirements.txt
│   ├── server.py
│   ├── voice2text_service.py
│   
├── model/
│   ├── model.py
│   
├── nginx/
│   ├── Dockerfile
│   ├── nginx.conf
│   
├── resources/
│   ├── down.wav
│   ├── I had some free time, so I wandered around town.mp3
│   ├── People living in town don't know the pleasures of country life.mp3
│   
├── client.py
└── docker-compose.yml

Description

Convert audio to text by calling this service. Ability to paraphrase the transcribed text using chatgpt paraphraser.

Transformers (Models Used)

The service utilizes pre-trained models for voice-to-text conversion. These models are based on the Hugging Face Transformers library and include:

1. Whisper for voice-to-text transformation
2. ChatGPT Paraphraser for paraphrasing

For details on the models' architecture and training, refer to the respective model documentation.

Setup Instructions

1. Clone the repository:

   git clone https://github.com/AdamLauz/voice2text.git
   cd voice2text

2. Install Docker: Follow the official Docker installation guide for your operating system.

3. Build Docker images and start containers:

docker-compose up --build

Usage

Once the Docker containers are running:

1. Use the provided client.py script to send audio files for transcription. The service supports both WAV and MP3 formats.

python client.py

2. Check the console output for the transcribed text.

API Endpoint

The API endpoint for transcribing audio files is:

http://localhost:80/predict

Architecture and How It Works

The web service architecture consists of three main components:

• Flask Server: Handles incoming HTTP requests, processes audio files, and invokes the voice-to-text service.
• Voice-to-Text Service: Utilizes trained models to convert audio input into text.
• Nginx Reverse Proxy: Routes requests from clients to the Flask server.

When a client sends an audio file to the /predict endpoint, the Flask server:

1. Saves the audio file locally.
2. Invokes the voice-to-text service to generate a transcription.
3. Sends back the transcribed text as a JSON response.

Example Call (Replace with Real Example)

Below is an example of how to call the service using cURL:

curl -X POST -F "file=@/path/to/your/audio/file.mp3" http://localhost:80/predict

Replace /path/to/your/audio/file.mp3 with the path to your actual audio file in MP3 format.

Additional Notes

• The Flask server is exposed on port 900 within the Docker network.
• Nginx is used as a reverse proxy to route requests to the Flask server.
• Make sure to provide audio files in WAV or MP3 format for transcription.

Troubleshooting

• If you encounter any issues, check the Docker logs for the Flask and Nginx containers.
• Ensure that the required dependencies are installed as specified in requirements.txt.

References

[1] Valerio Velardo. https://github.com/musikalkemist/Deep-Learning-Audio-Application-From-Design-to-Deployment