drmhse/nano-edge-tts: A 1.75M-Parameter Harmonic-plus-Noise Decoder for Kokoro-82M

huggingface.co2 min read

Researchers introduced nano-edge-tts, a 1.75-million-parameter harmonic-plus-noise decoder designed to replace Kokoro-82M's larger waveform decoder. By running two small networks at audio rate to set per-frequency gains on a pulse train and white noise, the model achieves a 33 times faster decoder speed on a single CPU thread while maintaining comparable output quality. Open article →

More on this topic