Audio Translator

A real-time audio transcription and translation system that captures audio streams and converts speech to text using OpenAI's Whisper model.

Features

• Real-time audio capture using PyAudio
• Stream-based speech recognition using Whisper
• Configurable transcription settings (language, model size)
• Callback system for real-time updates
• Efficient memory and CPU usage

Installation

1. Clone this repository:

git clone https://github.com/yourusername/audio_translator.git
cd audio_translator

2. Create a virtual environment and activate it:

python -m venv venv
# On Windows
venv\Scripts\activate
# On macOS/Linux
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

Dependencies

• Python 3.8+
• PyAudio
• NumPy
• OpenAI Whisper
• PyTorch
• Threading

Usage

Stream Recording Demo

To run the real-time transcription demo:

python audio_recorder/demo_stream.py --model tiny --language en --duration 60

Parameters:

• --model: Whisper model size (tiny, base, small, medium, large)
• --language: Language code for transcription (en, pt, es, etc.)
• --device: Audio device index (optional)
• --duration: Recording duration in seconds (default: 30)

Using the StreamRecorder in Your Code

from audio_recorder.src.whisper_transcriber import WhisperTranscriber
from audio_recorder.src.stream_recorder import StreamRecorder

# Initialize transcriber
transcriber = WhisperTranscriber(
    model_size="tiny",
    language="en",
    device="cuda" if WhisperTranscriber.is_cuda_available() else "cpu"
)

# Initialize recorder
recorder = StreamRecorder(
    transcriber=transcriber,
    device_index=None,  # Auto-select microphone
    callback_interval=0.5  # Update every 0.5 seconds
)

# Add callback for real-time updates
def on_transcription(text):
    print(f"Transcription: {text}")

recorder.add_transcription_callback(on_transcription)

# Start recording
recorder.start()

# Record for 10 seconds
import time
time.sleep(10)

# Stop recording
recorder.stop()

# Clean up
del recorder
del transcriber

Project Structure

audio_translator/
├── audio_recorder/
│   ├── src/
│   │   ├── __init__.py
│   │   ├── stream_recorder.py  # Real-time audio capture
│   │   └── whisper_transcriber.py  # Speech-to-text using Whisper
│   └── demo_stream.py  # Demo application
├── requirements.txt
└── README.md

License

MIT

Credits

This project uses OpenAI's Whisper model for speech recognition.