whatmodel.sh / tts / VALL-E X (Plachtaa open implementation)

VALL-E X (Plachtaa open implementation)

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

VALL-E X (Plachtaa open implementation) is an open-weights text-to-speech model from Plachtaa (community), released August 2023. Licence: mit.

Catalogue identity notes

Community reimplementation of Microsoft's VALL-E X at github.com/Plachtaa/VALL-E-X (vallex-checkpoint.pt); ClonEval names it as the open-source implementation it ran. Microsoft's VALL-E X was never released.

31.4±4.2#102 of 118 text-to-speech modelsfrom 5 benchmark results across 3 sources

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 VALL-E X (Plachtaa open implementation), out of 148 with a score.

#modelscorewhereRAMreleased
125Sesame CSM-1B Sesame34±5api $7.00/M chars +local~6.6GB2025-03
126VoiceCraft 830M (giga830M) Puyuan Peng et al. (UT Austin)32±4local~4.5GB2024-03
127VALL-E X (Plachtaa open implementation) Plachtaa (community)31±4local~1.6GB2023-08
128OuteTTS 0.2 500M OuteAI31±6local~0.9GB2024-11
129Simple-TTS ASAPP Research (Justin Lovelace et al.)31±5comparison only2023-10

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

$ what it's good at

other tests5 not in a task · 93% of its score
testresultvs the others in its rundifficultycountswhy
VERSA TTS demo · LibriSpeech test-clean UTMOS3.38 / 5 ✎▄·····▄·▄█·▄ beats 1/5easy · avg 4.11 / 5▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 45%at its level
ClonEval · LibriSpeech test-clean voice-clone speaker similarity (WavLM)90.1% ✎█·█········█ beats 2/2very easy · avg 97%▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 36%small run
CodecMOS-Accent VCTK voice clone · human speaker-similarity MOS (S-SPK-SIM)3.86 / 5 ✎▄▄█·▄▄▄▄▄·▄█ beats 4/11easy · avg 4.20 / 5▇▇ 5.7%blunt · at its level
CodecMOS-Accent VCTK voice clone · human naturalness MOS (S-NAT)3.07 / 5 ✎█··▄▄██▄·▄▄▄ beats 0/11easy · avg 3.99 / 5▇▇ 5.1%blunt · at its level
VERSA TTS demo · LibriSpeech test-clean Whisper WER27.3% ✎▄··········█ beats 1/5avg 6.4%▇ 1.2%blunt · noisy

$ where to run it

Other ways to run it

  1. Run it as published · needs ~1.62 GB · from github.com/Plachtaa (official)
    git clone https://github.com/Plachtaa/VALL-E-X.git && cd VALL-E-X && pip install -r requirements.txt && python -c "from utils.generation import SAMPLE_RATE, generate_audio, preload_models; from scipy.io.wavfile import write as write_wav; preload_models(); audio_array = generate_audio('Hello, my name is Nose. And uh, and I like hamburger. Hahaha... But I also have other interests such as playing tactic toast.'); write_wav('vallex_generation.wav', SAMPLE_RATE, audio_array)"

Official weights and documentation: huggingface.co/Plachta ↗

$ how the score was worked out · VERSA toolkit paper: TTS demo… 46%, ClonEval · Open Voice Cloning… 36%, CodecMOS-Accent: listening-te… 11%

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
VERSA toolkit paper: TTS demonstration (arXiv:2412.17667v2)2▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 46%CC BY 4.0
ClonEval · Open Voice Cloning Leaderboard (AMU)1▇▇▇▇▇▇▇▇▇▇▇ 36%Apache-2.0
CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1)2▇▇▇ 11%CC BY 4.0
prior · pull toward the middle▇▇ 7%
weighted average= 31.4

$ 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