Voice ID System

Voice ID System is a Python-based speaker recognition system that allows you to enroll your voice and later identify it from new recordings. It leverages modern speaker embedding techniques using Resemblyzer to extract robust voice features. The system is designed in a modular way for future enhancements and supports CPU, NVIDIA GPU, and AMD GPU configurations.

Features

• Enrollment Mode: Record a sample of your voice, extract a speaker embedding, and save it as your voice imprint.
• Identification Mode: Record a new sample and compare its embedding with the enrolled voice imprint using cosine similarity.
• Device Auto-Selection: Automatically selects a working audio input device or allows manual selection via a command-line option.
• CPU/GPU Support: Runs on CPU by default; supports NVIDIA and AMD GPUs if available.
• Modular Design: Easily extend or enhance the system (e.g., integrate advanced deep learning models) with separate modules.

Project Structure

voice-id-system/
 ├── main.py               # Main entry point for enrollment and identification.
 ├── audio_recorder.py     # Module for recording audio and selecting the input device.
 ├── voice_imprint.py      # Module for extracting and comparing voice embeddings.
 ├── model.py              # Placeholder for future deep learning model integration.
 └── requirements.txt      # Python package dependencies.

Requirements

Python Packages

This project requires Python 3.8 or higher and the following Python packages:

• torch (for GPU support, if applicable)
• torchaudio
• librosa
• numpy
• sounddevice
• scipy
• resemblyzer

Install them via pip using the provided requirements.txt file:

pip install -r requirements.txt

System Dependencies (Linux)

For audio recording support, install the necessary system libraries. On Ubuntu/Pop!_OS (or other Debian-based systems), run:

sudo apt-get install portaudio19-dev libasound-dev

If your system uses Pipewire (default on many modern distros), ensure you have the compatibility packages installed:

sudo apt install pipewire-alsa

How It Works

1. Enrollment Mode (--mode enroll):

   • Records your voice sample using your default (or a specified) audio input device.
   • Saves the raw audio as recording.wav (for reference).
   • Extracts a speaker embedding from the audio using Resemblyzer.
   • Saves the embedding (your voice imprint) to voice_imprint.npy.

2. Identification Mode (--mode identify):

   • Records a new audio sample.
   • Saves the sample as recording.wav.
   • Extracts a speaker embedding from the new recording.
   • Loads the saved voice imprint from voice_imprint.npy.
   • Compares the two embeddings using cosine similarity.
   • Reports whether the voice is identified based on a configurable threshold (default is 0.75).

3. Device Selection:

   • The system first attempts to use the default audio input device.
   • If the default device fails, it scans for available working devices and prompts you to choose one.
   • You can explicitly specify an audio device index with the --audio-device option.

4. CPU, NVIDIA, AMD Support:

   • The program runs on CPU by default.
   • Use the --device option (cpu, nvidia, or amd) to indicate your preferred computing device. The system will check GPU availability using PyTorch and fall back to CPU if necessary.

Installation

1. Clone the repository:

   git clone https://github.com/subhashdasyam/voice-id-system.git
   cd voice-id-system

2. Install Python dependencies:

   pip install -r requirements.txt

3. Install system dependencies (Linux):

   sudo apt-get install portaudio19-dev libasound-dev

   And if using Pipewire:

   sudo apt install pipewire-alsa

Usage

Enroll Your Voice

Enroll your voice to create a voice imprint. This command records 10 seconds of audio, extracts the speaker embedding, and saves it:

python main.py --mode enroll --duration 10 --device nvidia

Note: If you need to specify a particular audio device, use the --audio-device flag (e.g., --audio-device 2).

Identify Your Voice

Identify or verify your voice by comparing a new recording with the enrolled imprint:

python main.py --mode identify --duration 5

The program records 5 seconds of audio, extracts the speaker embedding, and compares it with the saved imprint using cosine similarity.

Troubleshooting

• Audio Device Issues:
  If you encounter errors with the default audio device, list available devices with:

  python -c "import sounddevice as sd; print(sd.query_devices())"

  Then, use the --audio-device option to specify a working device index.

• Permission Issues:
  Ensure your user account has the necessary permissions to access audio devices.

• Threshold Tuning:
  If identification results are not as expected (false accepts or rejects), adjust the cosine similarity threshold in voice_imprint.py within the compare_imprint function.

• GPU vs. CPU:
  The --device option only affects PyTorch device selection. If GPU is not available, the system automatically falls back to CPU.

Example

Future Enhancements

• Improved Speaker Verification Models:
  Integrate advanced deep learning models (e.g., x-vectors, d-vectors) for more robust recognition.
• Multiple Enrollment Samples:
  Allow multiple recordings during enrollment to create a more robust voice imprint.
• Cross-Platform Testing:
  Expand testing and optimization for Windows and macOS.

License

MIT

Acknowledgements

• Resemblyzer: For providing a robust framework for speaker embedding extraction.
• Open-source Community: For the excellent libraries and tools that make this project possible.