Datasets:
Add official standalone scorer + rewrite README (robustness, SEO, ZeroTTS links)
#3
by zeroweightai - opened
zerobench_eval/: scores pre-generated wavs, never loads a TTS model.
manifest -> synthesize -> score, in three commands.
README: why the WER policy is built this way (two ASRs, per-span reference
expansion, tone-only variants never admitted), how to score your own model,
current results, and links to ZeroTTS.
minhanh29 changed pull request status to merged