DeepCW: Real-time Neural Network Morse Code Decoder

Operating in challenging RF environments, DeepCW provides real-time Morse code decoding capabilities through a neural network model. The software excels at robustly interpreting weak signals, mitigating the effects of QSB, and performing effectively in noisy conditions, a common issue for CW operators. It also supports multi-channel decoding, allowing for the simultaneous interpretation of multiple CW signals present in the audio stream.

The application integrates audio pass-through functionality, enhanced with deep-learning-based noise reduction to clean up the incoming audio before decoding. Benchmarking data indicates a 0.00% error rate from 0 to -4 dB SNR, with errors remaining below 1.5% at -8 dB SNR and under 8% at -10 dB SNR across the full range of CW speeds. The project provides comparative analyses against established decoders such as CW Skimmer, fldigi, and ggmorse, utilizing publicly available CW QSO videos from YouTube to demonstrate its performance.

DeepCW is designed for cross-platform deployment, supporting Windows, macOS, Android, and iOS operating systems. The core DeepCW Engine, which houses the decoding model and a reference implementation, is available separately, offering Python and Node.js examples for decoding from WAV audio files.