v0.15.5

Highlights

• Download stack rebuilt (#765). The monolithic DownloadUtils is replaced by a composable ModelHub with resumable downloads (HTTP Range), byte-level progress, artifact validation before caching, and a shared retry policy that honors Retry-After. Breaking: DownloadUtils → ModelHub (#779).
• Parakeet unified ASR (#705). Native-Swift mel front-end, word-level timestamps, and lower-latency streaming tiers.
• Custom vocabulary controls (#647, #702, #724). Per-term CTC thresholds plus opt-in knobs to curb short-term over-firing and disable the acoustic spotter-rescue.

ASR

• Native-Swift mel front-end fixes iPadOS cold-start zero output (#744)
• Repair stale/corrupt Nemotron multilingual latin tokenizer on download (#690)
• Remove chunk-merge seam artifacts in offline transcription (#708), plus word-boundary-safe fallbacks for 3 residual seam-merge drop paths (#759)
• Fetch parakeet_vocab.json in AsrModels.download (#763)
• Feature Provider classes for unified Parakeet ASR (#713, @SGD2718)
• Fix type-checker timeout in NemotronMultilingualFleursBenchmark (#732, @kripskroll)

TTS

• Shared English text normalization (numbers/times) + initialism helpers (#715); adopted by StyleTTS2 (#720)
• Auto-chunk long text in KokoroAne high-level synthesize (#717)
• Fix KokoroAne strided MLMultiArray handling (#730, @XUJiahua)
• Supertonic: inter-chunk silence 0.3s → 0.05s + --silence flag (#737)

Diarization

• BNNS-fixed Sortformer v3 models, precision/compute-unit control + cancellation (#728)
• Deterministic, robust offline VBx re-clustering via K-Means n_init (#735, @testfields)
• Honor configuration.computeUnits in OfflineDiarizerModels.load (#743, @chadneal)
• Optional progressHandler on performCompleteDiarization (#753, @thatbin)
• Honor configuration.computeUnits in OfflineDiarizerModels.load (#743, @chadneal)
• Optional progressHandler on performCompleteDiarization (#753, @thatbin)
• Re-embed zero-vote spans instead of arbitrary cluster-0 tie-break (#751, @ComicBit)
• LS-EEND speaker pre-enrollment bugfixes (#729, @SGD2718)

Download infrastructure (#765)

• Wave 1: characterization tests + download smoke CI (#770)
• Wave 2: extract HFClient + RetryPolicy primitives (#771)
• Wave 3: one paginating HFTreeLister (#772)
• Wave 4: FileDownloader, ModelCache, ProgressReporter; DownloadUtils becomes a façade (#777)
• Wave 5: converge fetchHuggingFaceFile onto the shared policy (#778)
• Wave 6: ModelHub replaces DownloadUtils (#779) — breaking
• Resumable downloads (#768), byte-level progress (#764), byte-weighted subdirectory progress (#769, @JulianPscheid), artifact validation (#741), preserve cache on cancelled first load (#749, @ComicBit)

VAD

• Update Silero VAD CoreML artifact to v6.2.1 (#734, @LemonCANDY42)

Benchmarks & Docs

• Fail loudly on missing AMI annotations instead of scoring placeholder ground truth (#754)
• Fetch AMI corpus from HuggingFace mirror, Edinburgh fallback (#767)
• Add evoglyph to the Showcase (#755, @lucasmccomb)

New Contributors

@kripskroll, @testfields, @lucasmccomb, @thatbin, @chadneal, @holovchenko

Full Changelog: v0.15.4...v0.15.5

v0.15.4 japanese kokoro

What's Changed

• fix(tts): treat KokoroAne quote delimiters as punctuation by @LemonCANDY42 in #696
• fix(tts/kokoro-ane): synthesis correctness + doc accuracy by in #697
• fix(tts/kokoro-ane): KokoroNoise v2 — atan2 phase fix (removes HF sharpness) in #700
• feat(asr): expose per-token timings from StreamingUnifiedAsrManager by @ComicBit in #701
• Make SenseVoice downloads precision-aware by @pHequals7 in #703
• feat(tts/kokoro-ane): Japanese variant + PyTorch-matched output level by in #699

New Contributors

• @LemonCANDY42 made their first contribution in #696
• @ComicBit made their first contribution in #701
• @pHequals7 made their first contribution in #703

Full Changelog: v0.15.3...v0.15.4

v0.15.3

What's Changed

• feat(tts): M5 benchmark re-baseline + Kokoro ane-tail-gpu (M5 fix) + Supertonic int4 default + PocketTTS v2.1 ANE docs in #666
• fix(kokoro): make the M5-safe routing the default (Kokoro works on M5 out of the box) in #671
• chore(asr): remove experimental Parakeet CTC zh-CN Mandarin model in #675
• Remove experimental Magpie multilingual TTS backend in #674
• chore(asr): remove experimental Qwen3 ASR backend in #676
• fix(download): retry transient per-file failures in downloadRepo by @JulianPscheid in #681
• feat(tts/pocket): ANE placements — rank-4 split-KV models (.ane) + MLState pipeline (.aneState) by @Alex-Wengg in #679
• fix(kokoro): route the Noise stage to GPU in the M5-safe preset (+~10% synth) in #677
• feat(asr/eou): opt-in fused decoder+joint_decision path (+7-9% RTFx, WER neutral-or-better) in #680
• feat(asr): expose per-token timings from Nemotron streaming ASR (English + multilingual) by @JulianPscheid in #673
• Add TypeWhisper to the FluidAudio showcase by @SeoFood in #685
• fix(asr): make SlidingWindowAsrConfig.default fit the model's 240k-sample input in #689
• fix(asr): splice long-form chunk merges on SentencePiece word boundaries in #688
• fix(tts): Misaki-lexicon-first English frontend for KokoroAne in #692
• feat(asr): Parakeet Unified 0.6B backend (chunked-attention streaming + offline batch) in #693

New Contributors

• @JulianPscheid made their first contribution in #681
• @SeoFood made their first contribution in #685
• @rcourtman made their first contribution in #684

Full Changelog: v0.15.2...v0.15.3

v0.15.2

What's Changed

• feat(tts/supertonic3): quantized + ANE-bucketed VectorEstimator in #664
• PocketTTS v2.1: fused flow decoder (ANE) + cond prefill + fp16 flowlm (~1.8× RTFx) in #665

Full Changelog: v0.15.1...v0.15.2

v0.15.1

What's Changed

• perf(asr): opt-in GPU encoder placement for Parakeet v3 (+~8% RTFx, WER-neutral) in #659
• feat(asr): English Nemotron streaming 2240ms tier + B1 fusion (+50% RTFx, WER-neutral) in #660

Full Changelog: v0.15.0...v0.15.1

v0.15.0 — Multistreamer

What's Changed

• feat(asr): SenseVoiceSmall CoreML backend (multilingual, non-autoregressive) in #648
• feat(asr): Paraformer-large (zh) CoreML backend (non-autoregressive, CIF) in #651
• fix(download): fetch repo-root vocab.json for Cohere Transcribe (#649)g in #650
• feat(download): add offline-only enforcement via DownloadUtils.enforceOffline by @leecrossley in #632
• feat(asr): Nemotron 3.5 ASR Streaming Multilingual 0.6B (40 locales, on-device CoreML/ANE) g in #657

New Contributors

• @leecrossley made their first contribution in #632

Full Changelog: v0.14.8...v0.15.0

v0.14.8

What's Changed

• fix(asr): honor caller modelNames in DownloadUtils for non-baseline files (#524) in #625
• Fix/word boost improvements by @smdesai in #634
• fix(cli): expose missing Parakeet langs, add Greek script support by @rpkyle in #637
• Add timestamp support to EoU Streaming by @Kavi-Gupta in #629
• fix(asr): reduce English drift on French recordings via token blocklist by @Matth-93 in #630
• feat(diarizer): expose per-chunk embeddings on DiarizationResult by @adamsro in #633
• TTS: store downloaded models in Application Support, not Caches (iOS) in #642
• PocketTTS: stride-aware conditioning extraction for cloned voices in #643
• Supertonic-3: built-in voice enum + on-demand voice download in #644

New Contributors

• @rpkyle made their first contribution in #637
• @Kavi-Gupta made their first contribution in #629
• @jonyoder made their first contribution in #639

Full Changelog: v0.14.7...v0.14.8

v0.14.7

What's Changed

• feat(tts/pockettts): add 5 native-language voices + slim language-pack downloads ~40% in #624
• feat(cli): expose all unexposed config fields across 6 commands by @barzhomi in #616
• fix(asr): add opt-in v3 no-mel decode arbitration by @vdt4534 in #604

New Contributors

• @gmickel made their first contribution in #618
• @vdt4534 made their first contribution in #604

Full Changelog: v0.14.6...v0.14.7

v0.14.6

What's Changed

• feat(tts): Supertonic-3 multilingual CoreML TTS in #617
• refactor(tts): async StyleTTS2 predict + drop non-native Magpie synthesizeStream in #589
• Make SpeakerManager a struct and de-async DiarizerManager by @panv-kw in #591
• feat(tts/magpie): warmup API for cold-start mitigation (#60 Track 2) in #595* Fixed LS-EEND Memory Leak + Updated Docs by @SGD2718 in #605
• fix(tts/pocket-tts): repair v1 voice cloning for pocket-tts 2.0.0 (#592) in #601
• Timestamping RTTN decoder by @SGD2718 in #608
• fix(asr/nemotron): seed cache_len=1 to avoid ios17.slice_by_index zero-shape warning (#607) in #609
• fix(tts/pockettts): normalize French text and preserve mid-sentence chunks (#584) in #606
• feat(asr): expose tdtConfig in SlidingWindowAsrConfig by @execsumo in #611
• deprecate: remove CosyVoice3 and mono Kokoro (#571) in #602
• Diarization progress by @nburns in #615

Full Changelog: v0.14.5...v0.14.6

v0.14.5

What's Changed

• feat(tts/magpie): nanocodec v1/v2/v3 + decoder_step ANE pin + dual-precision API n #580
• feat(tts/magpie): nanocodec v4 (fp32 + int8 palettize) precision in #581
• fix(tts): guard direct Float16 reads with #if arch(arm64) (CosyVoice3, StyleTTS2) by @Beingpax in #582
• Add Utter app to showcase in README.md by @joepetrakovich in #585
• Fix: Prevent Metal crash when targetTokens is 0 in Kokoro TTS by @gregyoung14 in #586
• feat(tts): StyleTTS2 LibriTTS (iteration_3) CoreML backend in #588

New Contributors

• @joepetrakovich made their first contribution in #585
• @gregyoung14 made their first contribution in #586

Full Changelog: v0.14.4...v0.14.5