This project provides a neural network for end-to-end speech denoising, implementing a WaveNet architecture. It is targeted at researchers and developers working on audio processing and speech enhancement, offering a pre-trained model for immediate use and clear instructions for training and inference.

How It Works

The project utilizes a WaveNet architecture, known for its effectiveness in modeling sequential data like audio. This approach allows for deep convolutional layers with increasing dilation rates, enabling the model to capture long-range dependencies in the audio signal without requiring recurrent connections. This design is advantageous for speech denoising as it can effectively learn the complex patterns of speech and noise.

Quick Start & Requirements

• Install via pip install -r requirements.txt.
• Requires Keras 1.2 and Theano 0.9.0. TensorFlow 1.2.0 is not supported due to dilation limitations.
• Theano flags optimizer=fast_compile and device=gpu are recommended for usage.
• Dataset: "Noisy speech database for training speech enhancement algorithms and TTS models" (NSDTSEA) must be downloaded and extracted to data/NSDTSEA.
• Official quick-start and usage examples are provided within the README.

Highlighted Details

• Offers a pre-trained model ready for inference.
• Supports speedup by increasing target_field_length for faster denoising.
• Detailed configuration options are available in config.md.
• Allows for SNR computation if clean input paths are provided during inference.

Maintenance & Community

No specific information on contributors, sponsorships, or community channels (like Discord/Slack) is provided in the README.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project has a strict dependency on older versions of Keras (1.2) and Theano (0.9.0), which may pose significant challenges for setup and compatibility with modern deep learning environments. TensorFlow 1.2.0 is explicitly stated as unsupported.