whatmodel.sh / tts / LightSpeech + Parallel WaveGAN

LightSpeech + Parallel WaveGAN

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

LightSpeech + Parallel WaveGAN is an open-weights text-to-speech model from Microsoft NeuralSpeech, released February 2021. Its published size is 0.0018B parameters.

Catalogue identity notes

Original author implementation with linked pretrained acoustic and vocoder weights. Legacy environment requires a multi-step setup. No benchmark score or current one-command run inferred.

Model information sources

· Parameter count ↗

LightSpeech acoustic model only (makers paper: 1.8M vs. 27M for FastSpeech 2). The Parallel WaveGAN vocoder of this two-stage system is not included.

No comparable score yet. No reviewed benchmark results yet.

$ how it compares

Scored text-to-speech models released closest to LightSpeech + Parallel WaveGAN, out of 148 with a score.

#modelscorewhereRAMreleased
81Azure Ameha Neural Microsoft50±16comparison only2021-10
82Azure Mekdes Neural Microsoft50±16comparison only2021-10
124VITS (VCTK multi-speaker) Kakao Enterprise (Jaehyeon Kim et al.)34±5local~0.2GB2021-06
135Glow-TTS + HiFi-GAN (VCTK, VITS study) VITS study (public implementations)28±6comparison only2021-06
142Tacotron 2 + HiFi-GAN (VCTK, VITS study) VITS study (public implementations)19±6comparison only2021-06

see all text-to-speech models · find the ones that fit your machine

$ where to run it

Other ways to run it

version

  1. Run it as published · needs ~0.04 GB, NVIDIA only · from github.com/microsoft (official)
    git clone https://github.com/microsoft/NeuralSpeech.git && cd NeuralSpeech/LightSpeech && SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL=True pip install -r requirements.txt && wget https://huggingface.co/neuralspeech/lightspeech/resolve/main/data.tgz && tar -xzf data.tgz && mkdir -p checkpoints/lightspeech wavegan_pretrained && wget -P checkpoints/lightspeech https://huggingface.co/neuralspeech/lightspeech/resolve/main/model_ckpt_steps_100000.ckpt && curl -L -o wavegan_pretrained/checkpoint-1000000steps.pkl 'https://drive.usercontent.google.com/download?id=1FAOTzrVRDi-M_OdtfHRC7tS00YxMbzGz&export=download&confirm=t' && curl -L -o wavegan_pretrained/config.yml 'https://drive.usercontent.google.com/download?id=1Vlp2uM-ljY3nBEc9rDnQ8Z63QKhdRyKs&export=download&confirm=t' && curl -L -o wavegan_pretrained/stats.h5 'https://drive.usercontent.google.com/download?id=1c2AsCT0V95HQkaFqHlwl4q2ODerNa2yD&export=download&confirm=t' && CUDA_VISIBLE_DEVICES=0 PYTHONPATH=. python tasks/lightspeech_inference.py --config configs/tts/lj/lightspeech.yaml --exp_name lightspeech --reset --inference_text inference_text.txt

Official weights and documentation: github.com/microsoft ↗

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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