r9y9/wavenet_vocoder

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

2376 stars · 493 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A PyTorch implementation of the WaveNet vocoder that generates high-quality raw speech waveforms conditioned on acoustic features like mel-spectrograms. It includes pre-trained models and fast inference via cached convolution states, and integrates with ESPnet for full text-to-speech pipelines.

Use cases

• generate raw audio waveforms from mel-spectrograms
• build a text-to-speech pipeline with Tacotron2 and a neural vocoder
• train a WaveNet vocoder on a custom speech dataset
• synthesize speech with speaker embeddings for multi-speaker TTS
• experiment with mixture of logistics audio modeling
• speed up vocoder inference with cached convolutions

When to choose

• you need a neural vocoder to convert mel-spectrograms into audio
• you want a PyTorch WaveNet implementation with pre-trained LJSpeech models
• you are integrating a vocoder into an ESPnet-based TTS system

When to avoid

• you need a full text-to-speech model that converts raw text directly to speech
• you need fast real-time vocoding on CPU
• you want an actively developed project with recent updates and support

Facets

library · maturity maintenance

machine-learning deep-learning audio-processing speech-recognition speech-processing machine-learning deep-learning python cross-platform wavenet vocoder text-to-speech pytorch mel-spectrogram pretrained-models speech-synthesis audio

Member repositories

For agents

markdown · JSON · MCP: product_card(name="r9y9/wavenet_vocoder")