// audio · open-mmlab/Amphion

Amphion

Amphion é um projeto de áudio de código aberto de open-mmlab (open-mmlab/Amphion no GitHub). Tem 10.305 estrelas no GitHub e 851 forks, é publicado sob a licença MIT e foi atualizado pela última vez no GitHub em 25 de março de 2026.

Open research toolkit for text-to-speech, voice conversion, singing voice and text-to-audio generation.

★ 10,3 mil estrelas⑂ 851 forksMITatualizado mar 2026

Ver no GitHub ↗Official site ↗

▦ visual: o README deste projeto não inclui nenhuma captura incorporável,
então aqui vai um marcador. O guia abaixo está completo mesmo assim.

// dados do projeto

Dados do projeto

Repositório: open-mmlab/Amphion
Estrelas no GitHub: 10.305
Forks: 851
Licença: MIT
Última atualização no GitHub: 25 de março de 2026
Categoria: Áudio
Tipo de ficha: Guia de instalação completo
Site oficial: Official site

// como funciona

Como funciona

Set up the environment
Clone the repo, create the pinned conda environment and run the dependency script - or pull the Docker image instead.
Pick a task and model
Choose your generation task (TTS, voice conversion, singing, text-to-audio) and a model line, such as MaskGCT or Vevo for zero-shot speech, from the per-task READMEs.
Prepare data or grab a checkpoint
Use Amphion's unified preprocessing for datasets like LibriTTS, LJSpeech or VCTK, or download a pretrained checkpoint from Hugging Face.
Train or generate, then evaluate
Run the model's training or inference recipe and score results with the built-in objective metrics (intelligibility, speaker similarity, FAD and more).

// recursos

Por que as pessoas gostam tanto

Many TTS architectures

FastSpeech2, VITS, VALL-E, NaturalSpeech2, Jets, the non-autoregressive MaskGCT and the controllable zero-shot Vevo family, with per-model code and checkpoints.

Voice, accent and singing conversion

Zero-shot voice conversion (Vevo, FACodec, Noro), accent conversion, and singing voice conversion/synthesis including the unified Vevo2 framework for speech and singing.

Text-to-audio and neural codecs

A latent-diffusion text-to-audio model, plus neural audio codecs such as DualCodec and FACodec for token-based generation.

Vocoder zoo

GAN vocoders (MelGAN, HiFi-GAN, BigVGAN), flow (WaveGlow), diffusion (DiffWave) and autoregressive (WaveNet, WaveRNN) vocoders under one roof.

Objective evaluation suite

F0 and energy modeling metrics, intelligibility (CER/WER), spectrogram distortion (FAD, MCD, PESQ, STOI) and speaker-similarity scoring.

Emilia dataset and visualizations

The 100k+ hour Emilia in-the-wild speech dataset with its Emilia-Pipe preprocessing pipeline, and interactive visualizations such as SingVisio that explain classic model internals.

// execute você mesmo

Execute você mesmo

Clone the repository
Clone Amphion and enter the project directory.
```
git clone https://github.com/open-mmlab/Amphion.git
cd Amphion
```
Create the conda environment
Create and activate the Python environment the README specifies (a Docker image is the documented alternative).
```
conda create --name amphion python=3.9.15
conda activate amphion
```
Install dependencies
Run the project's environment script to install the required Python packages.
```
sh env.sh
```

// casos de uso

O que você pode criar

A PhD student reproducing a published TTS result before building on it.
Comparing MaskGCT, VITS and FastSpeech2 on the same dataset with shared metrics.
Building a singing-voice-conversion experiment with content features from Whisper or ContentVec.
Preparing a large in-the-wild speech corpus with Emilia-Pipe for training a new model.

// perguntas frequentes

Perguntas frequentes

Who maintains Amphion?

Amphion is maintained by the open-mmlab community and contributing research groups, with releases announced through the repository and its Discord.

Is Amphion free to use?

The toolkit is MIT-licensed. Individual datasets and some pretrained assets keep their own licenses - for example the Emilia dataset is CC BY-NC 4.0 - so check the asset you use.

Does it need a paid API key?

No. Amphion is a local research toolkit; models train and run on your own hardware, and you will want a GPU for serious training.

How does Amphion compare with Coqui TTS or ESPnet?

Those focus mainly on speech synthesis pipelines; Amphion covers a wider research surface - TTS, voice and singing conversion, text-to-audio, codecs, vocoders, metrics and datasets - with an emphasis on reproducing papers.

How do I get started?

Os outros projetos de Áudio com mais estrelas no catálogo do LibreAI — a lista realista de alternativas se Amphion não for o ideal para você.

MoneyPrinterTurbo★ 128,3 mil
利用 AI 大模型和自动化工作流，根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
Whisper★ 109,9 mil
Robust Speech Recognition via Large-Scale Weak Supervision
GPT-SoVITS★ 62,3 mil
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
Real-Time-Voice-Cloning★ 60,2 mil
Clone a voice from 5 seconds of audio and generate arbitrary speech in real time
whisper.cpp★ 54,1 mil
Port of OpenAI's Whisper model in C/C++

// mais em audio

Mais em Áudio

Projetos de Áudio próximos no catálogo, os mais próximos em estrelas de Amphion: continue explorando dos dois lados dele.

silero-vad★ 10,3 mil
Silero VAD: pre-trained enterprise-grade Voice Activity Detector
Mozilla TTS★ 10,2 mil
Mozilla's deep-learning text-to-speech library with Tacotron, Glow-TTS and neural vocoders.
pyannote-audio★ 10,6 mil
Neural building blocks for speaker diarization: speech activity detection, speaker change detection, overlapped speech detection, speaker embedding
RealtimeSTT★ 10,2 mil
A robust, efficient, low-latency speech-to-text library with advanced voice activity detection, wake word activation and instant transcription.
xiaozhi-esp32-server★ 10,8 mil
Self-hosted backend that gives ESP32 'Xiaozhi' devices streaming voice chat, memory and tool control.
espnet★ 10,0 mil
End-to-End Speech Processing Toolkit

Baixe o Amphion

Grátis e aberto: pegue o código-fonte, dê uma estrela no repositório e comece a criar.

Ver no GitHub ↗← Mais ferramentas de Áudio