whatmodel.sh / tts / FastSpeech 2 + HiFi-GAN (SpeechBrain LJ Speech)
FastSpeech 2 + HiFi-GAN (SpeechBrain LJ Speech)
↗ share
+ contribute
FastSpeech 2 + HiFi-GAN (SpeechBrain LJ Speech) is an open-weights text-to-speech model from SpeechBrain, released November 2022. Licence: apache-2.0.
Catalogue identity notes
SpeechBrain-trained FastSpeech2 acoustic model plus its LJ Speech HiFi-GAN vocoder. Original-paper MOS is not transferred to this checkpoint. CPU run command adapted from model card; memory is an estimated 4GB runtime budget, not a local measurement.
Model information sources
$ how it compares
Scored text-to-speech models released closest to FastSpeech 2 + HiFi-GAN (SpeechBrain LJ Speech), out of 148 with a score.
| # | model | score | where |
|---|---|---|---|
| 85 |  MMS TTS Amharic | 50±16 | local |
| 96 |  Piper (en_US lessac medium) | 47±5 | local |
| 132 | Tortoise TTS v2 | 29±2 | local |
| 139 |  Bark | 25±3 | local |
| 140 |  SpeechT5 TTS | 23±3 | local |
see all text-to-speech models · find the ones that fit your machine
$ where to run it
Other ways to run it
- Run it as published · needs ~4 GB · from huggingface.co/speechbrain (official)
uv run --with speechbrain --with soundfile python -c "import soundfile as sf; from speechbrain.inference.TTS import FastSpeech2; from speechbrain.inference.vocoders import HIFIGAN; t=FastSpeech2.from_hparams(source='speechbrain/tts-fastspeech2-ljspeech', savedir='pretrained/fastspeech2'); v=HIFIGAN.from_hparams(source='speechbrain/tts-hifigan-ljspeech', savedir='pretrained/hifigan'); mel,_,_,_=t.encode_text(['Hello world.']); wav=v.decode_batch(mel); sf.write('speech.wav', wav.squeeze().detach().cpu().numpy(), 22050)"
Official weights and documentation: huggingface.co/speechbrain ↗
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json