whatmodel.sh / tts / HierSpeech++ (LibriTTS-960)

HierSpeech++ (LibriTTS-960)

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

HierSpeech++ (LibriTTS-960) is an open-weights text-to-speech model from Korea University (Sang-Hoon Lee et al.), released November 2023. Its published size is 0.097B parameters. Licence: mit.

Catalogue identity notes

HierSpeech++ checkpoint trained on LibriTTS train-960 (README checkpoint table: 97M parameters, 555 h, 2,311 speakers). Distinct from the LibriTTS-460 and the Large v1/v1.1 English-Korean checkpoints (hierspeechpp-v1-1).

27.5±4.9#109 of 118 text-to-speech modelsfrom 3 benchmark results across 1 source

Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.

$ how it compares

The text-to-speech models ranked either side of HierSpeech++ (LibriTTS-960), out of 148 with a score, then more from Korea University (Sang-Hoon Lee et al.).

#modelscorewhereRAMreleased
134WhisperSpeech v1.95 Medium (7-language) Collabora28±3local~3.2GB2024-05
135Glow-TTS + HiFi-GAN (VCTK, VITS study) VITS study (public implementations)28±6comparison only2021-06
136HierSpeech++ (LibriTTS-960) Korea University (Sang-Hoon Lee et al.)28±5local~2.1GB2023-11
137Piper (tr_TR dfki medium) Open Home Foundation (Piper)26±4local~0.1GB2023-06
138OpenVoice V2 MyShell26±3local~0.8GB2024-04
more from Korea University (Sang-Hoon Lee et al.)
111HierSpeech++ v1.1 Korea University (Sang-Hoon Lee et al.)39±3local~2.1GB2023-11

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

$ what it's good at

other tests3 not in a task · 90% of its score
testresultvs the others in its rundifficultycountswhy
StyleTTS-ZS LibriSpeech-40 listening test · naturalness MOS3.54 / 5 ✎█·█········█ beats 0/2very easy · avg 4.64 / 5▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 42%at its level · small run
StyleTTS-ZS LibriSpeech-40 · HuBERT-large WER7.8% ✎██·········█ beats 0/2avg 0.6%▇▇▇▇▇▇▇▇▇▇▇ 28%noisy · small run
StyleTTS-ZS LibriSpeech-40 listening test · speaker-similarity MOS4.27 / 5 ✎█··█·······█ beats 2/2very easy · avg 4.55 / 5▇▇▇▇▇▇▇▇ 21%at its level · small run

$ where to run it

Other ways to run it

  1. Run it as published · needs ~2.1 GB, NVIDIA only · from github.com/sh-lee-prml
    git clone https://github.com/sh-lee-prml/HierSpeechpp && cd HierSpeechpp && pip install -r requirements.txt && pip install phonemizer && sudo apt-get install espeak-ng && CUDA_VISIBLE_DEVICES=0 python3 inference.py --ckpt "logs/hierspeechpp_libritts960/hierspeechpp_lt960_ckpt.pth" --ckpt_text2w2v "logs/ttv_libritts_v1/ttv_lt960_ckpt.pth" --output_dir "tts_results_eng_kor_v2" --noise_scale_vc "0.333" --noise_scale_ttv "0.333" --denoise_ratio "0"

Weights and documentation: github.com/sh-lee-prml ↗

$ how the score was worked out · StyleTTS-ZS: Efficient High-Q… 90%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
StyleTTS-ZS: Efficient High-Quality Zero-Shot Text-to-Speech Synthesis with Distilled Time-Varying Style Diffusion (arXiv:2409.10058v1)3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 90%CC BY-SA 4.0
prior · pull toward the middle▇▇▇ 10%
weighted average= 27.5

$ 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