flutter_edge_ai_speech 0.5.11

On-device speech (STT + TTS) for flutter_edge_ai via the LiteRT C API + dart:ffi. Opt-in Stt/TtsBackendProvider.

0.5.11

• A closed recognizer or synthesizer fails through the returned Future (#718).

0.5.10

• Docs: the macOS app setup STT and TTS need (#689).

0.5.9

• README: which engines default supportAudio from the model.

0.5.8

• Docs: after a barge-in, earlier audio turns stay in context with flutter_edge_ai_litertlm 1.15.0.

0.5.7

• VoiceSession.runAudioTurn: the model hears the recording, no speech-to-text.

0.5.6

• Require flutter_edge_ai ^2.3.0 and flutter_edge_ai_litertlm ^1.13.0.

0.5.5

• Closing a speech recognizer or synthesizer mid-request no longer leaks its native model.

0.5.4

• Require flutter_edge_ai ^2.1.0; README covers TTS from assets, files and bundles.

0.5.3

• Require flutter_edge_ai 2.x.

0.5.2

• Renamed from flutter_gemma_speech.
• Inflect TTS: speech was garbled — the encoder now gets the blank tokens it was trained with.

0.5.1

• Windows: STT/TTS failed with CreateTensorBufferFromHostMemory status 3 (needs flutter_gemma_litertlm 1.7.0).

0.5.0

• Whisper: output language is a parameter, no longer hardcoded English (#500).

0.4.3

• Add VoiceSession(streamAudio:) — clause-by-clause TTS overlapped with the LLM stream for lower time-to-first-audio.

0.4.2

• Add Inflect-Nano-v2 — fast on-device TTS (~90× real-time on CPU, English-only).
• VoiceSession.fromChat: optional onToolCall runs function calling in the Voice Loop (via core's generateChatResponseWithTools).

0.4.1

• Add Qwen3-TTS (multilingual AR codec-LM, 11 languages) — 2nd selectable TTS family.

0.4.0

• feat: on-device Whisper-tiny STT (English-only) — log-mel frontend + GPT-2 BPE decode.
• feat: on-device Parakeet-CTC STT (desktop) — NeMo mel frontend + greedy CTC decode.

0.3.0

• Add VoiceSession — on-device push-to-talk voice loop (STT → LLM → TTS) with barge-in.
• feat: robust TTS text frontend — punctuation-as-symbols, numbers/acronyms, neural OOV G2P, clause chunking.
• fix: model-agnostic TTS chunking — word-boundary + duration-aware split so long replies fit MAX_TEXT/MAX_MEL.

0.2.0

• Add on-device TTS (Matcha): installTts/getActiveTts, selectable model, PCM output.

0.1.0

• feat: on-device STT (moonshine-tiny) via LiteRtSttBackend + a generic, selectable SttModelProfile pipeline (encode → greedy decode → HF detokenize) on the shared LiteRT engine.