# kan-bayashi/ParallelWaveGAN

Unofficial Parallel WaveGAN (+ MelGAN & Multi-band MelGAN & HiFi-GAN & StyleMelGAN) with Pytorch

Repository: https://github.com/kan-bayashi/ParallelWaveGAN
Canonical: https://ross.abutalabs.com/products/parallelwavegan
Homepage: https://kan-bayashi.github.io/ParallelWaveGAN/
Language: Jupyter Notebook
License: MIT
License Family: permissive
Topics: speech-synthesis, neural-vocoder, text-to-speech, pytorch, wavenet, parallel-wavenet, realtime, tts, melgan, vocoder, hifigan, style-melgan
Last push: 2024-04-22T02:36:29+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 2501, "days_push": 864, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1645, forks 353 (observed 2026-08-28T04:05:16.248941+00:00)

## What it is
Unofficial PyTorch implementations of non-autoregressive neural vocoders including Parallel WaveGAN, MelGAN, Multi-band MelGAN, HiFi-GAN, and StyleMelGAN, with pretrained models and training recipes. It is designed for real-time speech synthesis and integrates with ESPnet-TTS pipelines.

## Use cases
- generate waveforms from mel spectrograms with a neural vocoder
- train a HiFi-GAN or Parallel WaveGAN vocoder on my own dataset
- build a real-time text-to-speech pipeline with ESPnet
- convert mel spectrograms to audio in real time
- fine-tune a pretrained vocoder for a new speaker or language
- train a singing voice vocoder

## When to choose
- you need PyTorch vocoder models compatible with ESPnet-TTS
- you want multiple state-of-the-art vocoder architectures in one codebase with pretrained checkpoints
- you need real-time non-autoregressive waveform generation on GPU

## When to avoid
- you need an officially maintained implementation or long-term support
- you want a vocoder outside the GAN-based mel-spectrogram family
- you need CPU-only real-time inference at production scale

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, audio-processing, speech-recognition, tts
- domain: speech-processing, machine-learning, deep-learning
- platform: python, cross-platform
- tags: vocoder, parallel-wavegan, melgan, hifi-gan, stylemelgan, pytorch, text-to-speech, neural-vocoder, pretrained-models, espnet, audio, gpu

## Member repositories
- kan-bayashi/ParallelWaveGAN (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:16.248941+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:45:25.283673+00:00, confidence not recorded.
  - readme: https://github.com/kan-bayashi/ParallelWaveGAN (fetched 2026-08-28T04:05:16.248941+00:00, sha 309b50f057b1)
  - homepage: https://kan-bayashi.github.io/ParallelWaveGAN/ (fetched 2026-08-29T11:18:53.703492+00:00, sha b94bc81c2c48)
- Data as of 2026-08-30T08:39:29.467469+00:00.