← Back to Home
Changelog
v0.5.1
2026-04-12Ultimate Optimization
- ● Fixed hotword format bug: v0.5.0 hotwords were silently ignored by ASR server, now using official format
- ● WebSocket warm-up: eliminates 200-500ms cold start delay, speak instantly on key press
- ● Early audio capture with buffering: no more lost initial audio
- ● Enabled macOS Voice Processing Mode: hardware-level noise cancellation + auto gain
- ● Pre-emphasis filter: clearer high-frequency consonants for better recognition
- ● Audio aggregation optimization: 70% reduction in frame overhead
- ● Added correction vocabulary + DDC spoken language smoothing
- ● Batch text insertion: 10-20x faster insertion, auto-fallback to clipboard for long text
- ● Context-aware LLM correction: includes recent records + current app name
v0.5.0
2026-04-12Accuracy Optimization
- ● Three-layer accuracy pipeline: hotword injection → local replacement → async LLM correction
- ● Hotword system: custom proper nouns for better recognition, zero latency
- ● Local replacement rules: remove filler words, fix casing, synchronous microsecond processing
- ● LLM correction: runs asynchronously in background, suggests fixes via menu bar notification
- ● Optimized ASR parameters: adjusted end_window_size to reduce sentence truncation
- ● New "Accuracy" tab in Settings with visual pipeline configuration
- ● Replacement rule presets: one-click loading of common rules
v0.4.0
2026-04-12AI Memory
- ● Added LLM service gateway with OpenAI-compatible API support
- ● AI auto-classification: LLM + local rules fallback dual engine
- ● Natural language queries: search voice history using natural language
- ● Daily briefings and weekly insights
- ● All AI calls run asynchronously, never blocking recording or input
v0.3.0
2026-04-11Data Center
- ● Voice input history auto-saved
- ● History search and filter functionality
- ● Data statistics and analysis
- ● SwiftData local storage
v0.2.0
2026-04-11Core MVP
- ● Global hotkey (right Option key) push-to-talk, release to input
- ● Volcano Engine streaming ASR integration with real-time Chinese recognition
- ● Auto-insert text at cursor position, works in all apps
- ● Menu bar presence with recording status indicator
- ● First-launch onboarding flow (permission authorization guide)
- ● Settings page: hotkey config, ASR parameters, API key management