v0.16.0 — CUA-S1-FORM

What's Changed

• Add CUA-S1-FORMS Core ML scoring and benchmarks in #936
• fix(asr/streaming): seam reconciliation decides by word text, not frames (#897 corpus follow-up) in #904
• fix(asr): resolve sentence-final punctuation ids from the loaded vocabulary (#905) in #906
• fix(asr/streaming): re-decode every window on a fresh state; end-align the final window (#897) in #908
• fix(asr): recover whole-window blank decodes with a length-perturbation ladder (#909) in #910
• fix(tts/kokoro-ane): stem possessives instead of G2P-ing the whole token by @akmittal006 in #884
• feat(tts/kokoro-ane): in-process Japanese text frontend (MeCab over trimmed unidic-lite + Cutlet rules) (#914) in #915
• feat(tts): Chatterbox backends — Multilingual + Nano (T3 + S3Gen CoreML chains) (#49) in #907
• fix(download): include required root-level files without a metadata extension in #918
• fix(cli/scoring): make TextNormalizer replacement order deterministic in #921
• fix(cli/scoring): sort speaker ids so diarization speaker mapping is deterministic in #923
• feat(tts/chatterbox-nano): extended ~30 s output capacity + honest budget errors in #925
• Load local Orukeet-compatible ASR bundles without repository fallback by @Nathan-Roll1 in #928
• fix(tts/pocket): keep cache-safe long sentences whole in #938
• Pin diarization artifacts to immutable revision (#927) in #939

New Contributors

• @akmittal006 made their first contribution in #884
• @Nathan-Roll1 made their first contribution in #928

Full Changelog: v0.15.7...v0.15.8