# jik876/hifi-gan

HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis

Repository: https://github.com/jik876/hifi-gan
Canonical: https://ross.abutalabs.com/products/hifi-gan
Language: Python
License: MIT
License Family: permissive
Topics: speech-synthesis, gan, text-to-speech, tts, deep-learning, hifi-gan, pytorch, vocoder
Last push: 2024-07-27T20:56:30+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2149, "days_push": 767, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2367, forks 557 (observed 2026-08-28T04:06:41.611624+00:00)

## What it is
The official PyTorch implementation of HiFi-GAN, a generative adversarial network that converts mel-spectrograms into high-fidelity 22.05 kHz audio faster than real time on a single GPU. It ships with training scripts, multiple generator configurations, and pretrained checkpoints for LJSpeech and VCTK, including versions fine-tuned for Tacotron2 and universal multi-speaker use.

## Use cases
- convert mel-spectrograms to audio waveforms
- add a vocoder to a text-to-speech pipeline
- generate high fidelity speech from spectrograms
- train a GAN vocoder on my own dataset
- download a pretrained vocoder model
- run real-time speech synthesis on GPU or CPU
- fine-tune a vocoder for Tacotron2 output

## When to choose
- You need a proven, high-quality vocoder with ready-to-use pretrained checkpoints (LJSpeech, VCTK, Tacotron2 fine-tuned)
- You want faster-than-real-time waveform generation, including a small-footprint model that runs 13x faster than real time on CPU
- You are reproducing the HiFi-GAN paper or building research on top of it
- You work in Python/PyTorch and want reference training and inference code

## When to avoid
- You need a complete end-to-end TTS system - this is a vocoder only, with no text normalization, phonemization, or acoustic model included
- You need ONNX, TensorFlow, or non-Python deployment support - the official code is PyTorch-only
- You require guaranteed ongoing maintenance or new features - the repo is a research artifact updated sporadically
- You need production streaming or low-latency chunked inference support out of the box

## Facets
- artifact type: library
- maturity: stable
- function: tts, audio-processing, machine-learning, deep-learning
- domain: speech-processing, deep-learning, machine-learning, artificial-intelligence
- platform: python, cli, cross-platform
- tags: vocoder, gan, speech-synthesis, mel-spectrogram-inversion, pytorch, pretrained-models, text-to-speech, research-implementation, audio, gpu

## Member repositories
- jik876/hifi-gan (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:41.611624+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 2026-08-30T02:35:57.667941+00:00, confidence not recorded.
  - readme: https://github.com/jik876/hifi-gan (fetched 2026-08-28T04:06:41.611624+00:00, sha 5fad2885568c)
- Data as of 2026-08-30T08:39:29.467469+00:00.