v1.1.1 — Warning cleanup

Fixes all compiler warnings across 6 files: unsafeBitCast->unsafeDowncast, a macOS 14 deprecation, an NSPasteboard/Sendable capture, and AVAudioConverterInputBlock concurrency false-positives. No behavior changes — 31/31 tests pass, verified with a from-scratch build.

Full Changelog: v1.1.0...v1.1.1

v1.1.0 — Dictation reliability fixes

Fixed

• Dictation state could get permanently stuck showing "recording" (mic icon, notch overlay) if starting the recognizer failed for any reason — the UI never reset, so it looked like it was listening while nothing was actually happening.
• Speech recognition defaulted to server-based (network) processing, which could silently fail to transcribe genuine, correctly-captured audio. Now uses on-device recognition when available, matching this package's on-device positioning.
• Added unified-logging diagnostics throughout the capture/recognition pipeline for real debuggability from Console.app / log stream.

Full Changelog: v1.0.0...v1.1.0

v1.0.0 — First standalone release

On-device voice dictation, command routing, and AI-assisted rewriting for macOS, as a standalone Swift package with no dependency on any host app.

What's included

• Local ASR via Apple's SFSpeechRecognizer/SpeechAnalyzer, or downloadable on-device models (Whisper, Parakeet, Nemotron, Cohere Transcribe) via CoreML
• Global hotkey activation — hold-to-record, toggle, or double-tap
• Command routing via a configurable wake word, and an AI-backed edit mode for rewriting selected text
• Notch/bottom overlay with live transcript and waveform
• History, stats, and custom dictionary, all local file-based persistence

This release

• Fixes a crash in every transcription provider when microphone permission isn't yet granted — all AVAudioEngine-based providers now check authorization before touching inputNode, surfacing a catchable error instead of a hard crash.

Example app

See the README for AICompleteChat, a full-source, on-device AI chat app that wires AiVoiceKit into DesignFoundation/DesignFoundationPro.

License

GPLv3 (derivative of FluidVoice by Aether AI Studio / altic-dev) — see LICENSE.