VALL-E 2
Microsoft's zero-shot TTS research model — claimed human parity, never released
About
VALL-E 2 (arXiv:2406.05370, June 2024) is Microsoft Research's follow-up to VALL-E, introducing repetition-aware sampling and grouped code modeling to fix the original's word-repetition and stability problems. The paper claims it is the first zero-shot text-to-speech system to reach human parity on LibriSpeech and VCTK across robustness, naturalness, and speaker similarity, cloning a voice from seconds of audio. Microsoft's release notes are explicit: VALL-E 2 is purely a research project with no plans for productization or public access, citing spoofing and impersonation risks. Treat the human-parity claim as team-reported on academic benchmarks; there are no weights, no API, and no independent reproduction path. Its value is as a reference point for how far zero-shot voice cloning has come — and why the frontier labs are keeping it locked up.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: WER, SMOS, and CMOS at or above ground truth on LibriSpeech and VCTK
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Microsoft Research paper arXiv:2406.05370, submitted June 8, 2024; Human-parity claims on LibriSpeech/VCTK are team-reported on academic benchmarks; Microsoft explicitly declined to productize; no weights or API — research only.