whatmodel.sh / tts / Tacotron 2 + WaveGlow (NVIDIA)

Tacotron 2 + WaveGlow (NVIDIA)

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

Tacotron 2 + WaveGlow (NVIDIA) is an open-weights text-to-speech model from NVIDIA, released 2018. Licence: bsd-3-clause.

Catalogue identity notes

Exact NVIDIA implementation with WaveGlow vocoder. Do not transfer these listening-study values to Coqui, SpeechBrain, HiFi-GAN or a differently trained checkpoint. Historical run dependencies documented upstream; no current one-command run verified.

Model information sources

· Weights licence ↗

No comparable score yet. Its published results are below, but they don't overlap enough with other models to calculate a score.

$ how it compares

Scored text-to-speech models released closest to Tacotron 2 + WaveGlow (NVIDIA), out of 148 with a score, then more from NVIDIA.

#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
more from NVIDIA
113Magpie TTS Multilingual 357M NVIDIA39±6local~0.7GB2026-07

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

$ what it's good at

other tests1 not in a task · 0.0% of its score
testresultvs the others in its rundifficultycountswhy
FastPitch 2020 LJ Speech human MOS lab3.95 / 5 ✎█··········█ beats 0/1—not counted yet: no other model in the same run to compare it with

$ where to run it

Other ways to run it

  1. Run it as published · needs ~1.19 GB, NVIDIA only · from github.com/pytorch
    pip install torch numpy scipy librosa unidecode inflect && apt-get update && apt-get install -y libsndfile1 && python -c "import torch; from scipy.io.wavfile import write; tacotron2 = torch.hub.load('NVIDIA/DeepLearningExamples:torchhub', 'nvidia_tacotron2', model_math='fp16', trust_repo=True); tacotron2 = tacotron2.to('cuda'); tacotron2.eval(); waveglow = torch.hub.load('NVIDIA/DeepLearningExamples:torchhub', 'nvidia_waveglow', model_math='fp16', trust_repo=True); waveglow = waveglow.remove_weightnorm(waveglow); waveglow = waveglow.to('cuda'); waveglow.eval(); utils = torch.hub.load('NVIDIA/DeepLearningExamples:torchhub', 'nvidia_tts_utils', trust_repo=True); sequences, lengths = utils.prepare_input_sequence(['Hello world, I missed you so much.']); torch.set_grad_enabled(False); mel, _, _ = tacotron2.infer(sequences, lengths); audio = waveglow.infer(mel); write('audio.wav', 22050, audio[0].data.cpu().numpy())"

Official weights and documentation: github.com/NVIDIA ↗

$ published results · 1 result, not enough overlap for a score
testresultsource and conditions
FastPitch 2020 LJ Speech human MOS3.95 / 5NVIDIA FastPitch: original listening studies in committed README · lab-reported

Committed README MOS table: 3.946 ± 0.134 (95% CI); grapheme inputs and WaveGlow (FastPitch paper Table 1). Store (MOS−1)/4.

$ 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