v0.1.3

What's Changed

• Fill in some missing codecs by @lucasnewman in #96
• Add Echo TTS by @lucasnewman in #97
• Add FireRed ASR 2 by @lucasnewman in #98
• Add SenseVoice (ASR/AER/LID) by @lucasnewman in #99
• Add Chatterbox Turbo by @lucasnewman in #102
• Update README.md by @Blaizzy in #104
• Add Fish Audio S2 Pro model by @lucasnewman in #106
• Add README for Qwen3 TTS by @lucasnewman in #107
• Fix Parakeet multilingual recognition (Russian/non-English) by @beshkenadze in #108
• Allow autodetecting language for Qwen3 ASR by @lucasnewman in #110
• feat: add MLXAudioG2P — neural ByT5 + dictionary lexicons by @beshkenadze in #121
• feat: add StyleTTS2 shared blocks, ALBERT encoder, and English G2P by @beshkenadze in #122
• feat: add KittenTTS text-to-speech model by @beshkenadze in #123
• feat: add Kokoro TTS with multilingual support by @beshkenadze in #124
• Add MLXFast dependency for MLXAudioCodecs by @noorbhatia in #128
• Add missing MLXFast dependency to MLXAudioSTT by @Hurryitup in #132
• Adding fromModelDirectory methods to various models by @ptsochantaris in #144
• fix: Kokoro 8-bit crash on quantized weight layout and NaN durations by @beshkenadze in #136
• Add Cohere Transcribe STT support by @beshkenadze in #129
• fix: Parakeet int8 Compatibility Fix by @RaihanulHaque in #145
• fix: Qwen3 TTS - handle mixed types in spkId and spkIsDialect for CustomVoice model by @Ichigo3766 in #146
• feat: speaker embedding support for Qwen3-TTS CustomVoice models by @Ichigo3766 in #148
• fix(cohere): load unmodified Python-produced quantized checkpoints by @beshkenadze in #153
• [Qwen3 TTS] Cache reference audio between generations by @lucasnewman in #125
• Add context support to Qwen3 ASR by @lucasnewman in #126
• fix: pass textProcessor through in KokoroModel.fromPretrained by @Oliver2213 in #151
• fix: exclude in-tree docs from SwiftPM targets by @gaelic-ghost in #152
• models: hide marvis rope caches from strict verification by @gaelic-ghost in #163
• models: escape underscore member reads in EnglishG2P by @gaelic-ghost in #161
• Upgrade package versions and adopt changes from mlx-lm-swift by @lucasnewman in #158
• Add DeepFilterNet speech enhancement model by @kylehowells in #100
• Parakeet(perf): batched decode, bf16 compute, TDT perf fixes by @beshkenadze in #167
• Add optional progressHandler to ModelUtils.resolveOrDownloadModel by @Oliver2213 in #166
• feat: add reusable qwen3 reference conditioning by @gaelic-ghost in #156
• docs: add contributing guide by @beshkenadze in #142
• Fix (Qwen3TTS): pass topK and minP from GenerateParameters by @Oliver2213 in #168
• Improved task cancellation handling for generation by @lucasnewman in #157
• Fix Qwen3ASR stream cancellation propagation by @summereasy in #172
• Add MOSS-TTS-Nano model by @lucasnewman in #169
• Add Silero VAD model by @beshkenadze in #176
• Qwen3-ASR: chunked prefill + asyncEval + repetition penalty + loop guard by @beshkenadze in #174
• Add MOSS-TTS model family by @lucasnewman in #179
• fix(MossTTS): gate homeDirectoryForCurrentUser behind #if os(macOS) by @brunobar79 in #189
• FireRedASR2Model: hide CMVN runtime arrays from strict-verify reflection by @heyzgj in #188
• perf(sortformer): single bulk readback in predsToSegments (~1.8x streaming) by @beshkenadze in #193
• Add Nemotron 3.5 ASR streaming model support by @vanch007 in #195
• Nemotron 3.5 ASR: fix prompt_kernel load crash + add cache-aware streaming (builds on #195) by @beshkenadze in #196
• refactor(stt): extract shared NeMo-family module; decouple NemotronASR from Parakeet (#197) by @beshkenadze in #198
• Kokoro: fix French nasal vowels dropped by Character tokenization by @slikmedia in #204
• Add Irodori-TTS (Japanese flow-matching TTS) by @jonpol01 in #206
• Add Whisper model family by @JacobLinCool in #192
• feat(voxtral): true incremental streaming for Voxtral Realtime by @beshkenadze in #205
• Fix Qwen3-TTS CustomVoice voice parsing by @summereasy in #186
• feat(nemotron_asr): incremental NemotronASRStreamSession for live mic by @beshkenadze in #208
• Voxtral Realtime: add Silero VAD pre-processor by @beshkenadze in #178
• Cohere Transcribe: add Silero VAD pre-processor by @beshkenadze in #177
• Add OmniVoice TTS (NAR diffusion over Qwen3 + HiggsAudioV2 codec, with voice cloning) by @JacobLinCool in #209
• docs(omnivoice): bf16 weights now include the complete tokenizer by @beshkenadze in #213
• Fix remote tokenizer loading for the MOSS models by @lucasnewman in #207
• Add Xcode project gen for SimpleChat example by @lucasnewman in #211
• Fill in remaining codecs for parity with mlx-audio by @lucasnewman in #214
• Add additional STT models by @lucasnewman in #215
• Add MOSS-Transcribe-Diarize by @lucasnewman in #221
• Add IndexTTS by @lucasnewman in #220
• Improve Voices STT controls and MOSS streaming by @Blaizzy in #223

New Contributors

• @noorbhatia made their first contribution in #128
• @Hurryitup made their first contribution in #132
• @ptsochantaris made their first contribution in #144
• @RaihanulHaque made their first contribution in #145
• @Ichigo3766 made their first contribution in #146
• @Oliver2213 made their first contribution in #151
• @gaelic-ghost made their first contribution in #152
• @kylehowells made their first contribution in #100
• @summereasy made their first contribution in #172
• @brunobar79 made their first contribution in #189
• @heyzgj made their first contribution in #188
• @vanch007 made their first contribution in #195
• @slikmedia made their first contribution in #204
• @jonpol01 made their first contribution in #206
• @JacobLinCool made their first contribution in #192

Full Changelog: v0.1.2...v0.1.3