v0.12.0 PocketTTS

New TTS model converted from the Pocket TTS

• simpler chunking algorithm while supporting longer token counts
• EOS detection stops generation naturally instead of hitting a fixed token wall
• real time support
• No espeak dependency
• iOS RAM friendly

Bug Fixes

• #282 — Custom vocabulary rescoring now applies to chunked long audio (was silently skipped on >15s
  audio) — thanks @Beingpax

Note

• we plan to deprecate kokoro tts in the future soon

v0.11.0 Custom Vocabulary & Pipeline Updates

What’s New

• Custom Vocabulary Support (#251): Major feature for custom vocab with recognized domain terms
• Vocabulary Pipeline Refactor (#276): Restructure, dead code removal, and pure Swift dataset
• download Float16 Xcode GUI Build Failure (#270): Add architecture checks to avoid build break by @schmatz
• Documentation Refresh (#280): Reorganize docs and remove stale content README Updates: Link and description cleanup

Full Changelog: v0.10.1...v0.11.0

v0.10.1: Streaming & De-esser

What's New

• Streaming Audio Processing (#257): Memory-efficient transcription for large files — 99.5% reduction (230MB → ~1.2MB constant)
• TTS De-esser (#267): Reduces harsh sibilant sounds in Kokoro TTS output, on by default

Bug Fixes

• macOS 26 Sortformer compatibility (#266): Switch to V2 models to fix BNNS compiler error
• Chunk boundary transcription loss (#264): Fix speech truncation by prepending mel context to non-first chunks
• Legacy FileHandle.write (#262): Replace deprecated API with throwing write(contentsOf:)

New Contributors

• @starkdmi — chunk boundary transcription fix (#264)

Full Changelog: v0.10.0...v0.10.1

v0.10.0: Sortformer

Sortformer: Real-Time Speaker Diarization

CoreML version of Nividia's Sortformer

• Real-time streaming - Speaker labels as audio comes in
• Noisy environment support - Works where traditional pipelines fail
• Overlapping speech - Scores all 4 speakers independently per frame, multiple can be active simultaneously
• Single neural model - No complex pipeline, just one model

---

Credit to @SGD2718 for the Sortformer implementation & model conversion.

v0.9.1

What's Changed

Bug Fixes

• fix: Swift 6 Sendable errors with macOS 26.2 SDK (#245) by @tacshi
• fix: Swift 6 concurrency errors in audio conversion (#239) by @Alex-Wengg
• fix: rename CLI executable to fluidaudiocli to avoid Xcode name collision by @Alex-Wengg
• fix(diarizer): use K-Means centroids when speaker count constraint is applied (#236) by @beshkenadze
• Preventing loops with non-blank tokens (#244) by @Steven-Weng

New Contributors

• @tacshi made their first contribution in #245

Full Changelog: v0.9.0...v0.9.1

v0.9.0 - Swift 6 Support

What's New

Swift 6 Support

• Full Swift 6 compatibility
• Updated swift-tools-version

Full Changelog: v0.8.2...v0.9.0

v0.8.2

What's Changed

• Padding blank audio to the back of short audio files Bug Fix (#234) - @Steven-Weng
• feat(tts): SSML tag phoneme, sub, and say-as support for kokoro TTS coreml (#235) - @smdesai

Full Changelog: v0.8.1...v0.8.2

v0.8.1

What's New

Features

• Transcription progress: Support emitting transcription progress percentage (#229) - @xinnjie
• Speaker count constraints: Added minSpeakers, maxSpeakers, numSpeakers options for diarization (#220) - @beshkenadze
• CLI JSON output: Added --output-json flag to transcribe command (#222) - @Steven-Weng

Bug Fixes

• Fixed segmentation prewarm hang using async prediction API (#227) - @Tasktivity
• Fixed content skipping bug (#223) - @Steven-Weng

New Contributors

• @xinnjie made their first contribution in #229
• @beshkenadze made their first contribution in #220
• @Tasktivity made their first contribution in #227
• @Steven-Weng made their first contribution in #222

Full Changelog: v0.8.0...v0.8.1

v0.8.0

What's New

Parakeet EOU Streaming ASR (#216)

New streaming ASR with End-of-Utterance (EOU) detection using NVIDIA's Parakeet EOU 120M model.

Features:

• StreamingEouAsrManager - streaming pipeline with 160ms and 320ms chunk support
• Real-time End-of-Utterance detection with configurable debounce (default 1280ms)
• Native Swift NeMoMelSpectrogram with vDSP vectorization
• RnntDecoder - RNN-T greedy decoder with EOU detection
• Automatic model downloads from HuggingFace

CLI:

swift run fluidaudio parakeet-eou audio.wav --chunk-ms 320

Full Changelog: v0.7.12...v0.8.0

v0.7.12

v0.7.12

• Custom TTS pronunciation dictionaries — lexicon file support for Kokoro (CLI --lexicon, TtsCustomLexicon, runtime updates), multi-tier matching, and docs/tests (#213) by smdesai
• Hugging Face downloads — integrate the official Swift Hugging Face SDK for model fetching with env token/proxy support and leaner registry traversal (#215)
• Platform/model validation utilities — add SystemInfo.isIntelMac, AsrModels.isModelValid, streaming decoder reuse, and non-contiguous stride handling (#210)