jarvis_dart 1.0.8

A Dart-based voice assistant application inspired by JARVIS from Iron Man.

1.0.8

• Fix example file to use correct ConfigLoader API

1.0.7

• Improve pub.dev analysis score (120 → 160 points):
  • Fix LICENSE file format for OSI recognition
  • Add example file for documentation
  • Update archive dependency to latest (4.0.7)
  • Fix code formatting and lint issues

1.0.6

• Fix audio buffering during follow-up state (prevents empty transcriptions)
• Improve sentence processing for faster TTS streaming:
  • Break on newlines for multi-line responses
  • Break on clause separators (; : — –)
  • Force break at 20 words max to prevent long pauses
• Enable Metal GPU acceleration for llama-cli (-ngl 99)
• Add VAD unit tests for follow-up listening scenarios

1.0.5

• Generate tuned keywords.txt on fresh install
• Add JARVIS phonetic variants for better wake word detection
• Always regenerate keywords.txt on setup to apply latest tuning
• Clean up Python venv after adding ONNX metadata (saves ~50MB)
• Generate acknowledgment audio during setup using Dart sherpa_onnx (no Python required)
• Auto-resolve audio assets from ~/.jarvis/assets when running globally
• Add timing instrumentation for performance analysis (--debug flag)
• Pipeline TTS synthesis with playback for smoother multi-sentence responses
• Generate complete config.yaml with all settings during setup
• Use isolate-based TTS for true parallel synthesis (~27% speedup)
• Add whisper-server support to keep whisper model warm between calls
• Stream LLM output to TTS for faster time-to-first-audio (reduced from 3-5s to ~1s)

1.0.4

• Add ONNX metadata step to TTS model setup
• Create Python venv automatically for onnx package
• Add sherpa-onnx required metadata (model_type, language, voice, sample_rate)
• Require Python 3.8+ with clear error messages

1.0.3

• Auto-detect whisper-cli and llama-cli executables
• Search common locations: /opt/homebrew/bin, /usr/local/bin, PATH
• Try multiple executable names (whisper-cli, whisper, llama-cli, llama)
• Show dynamic "Next steps" based on detected tools

1.0.2

• Fix platform-specific sherpa library detection
• Detect correct platform: macOS, Linux, or Windows
• Find sherpa_onnx_macos, sherpa_onnx_linux, or sherpa_onnx_windows
• Select latest version when multiple versions installed

1.0.1

• Add whisper model download to jarvis setup
• Download ggml-base.en.bin (~142MB) from HuggingFace
• Store in ~/.jarvis/models/whisper/
• Auto-configure whisper_model_path in generated config

1.0.0

• Initial release
• Global CLI install via dart pub global activate jarvis_dart
• jarvis setup command for first-time model download
• Wake word detection using sherpa-onnx
• Speech-to-text using whisper.cpp
• LLM responses using llama.cpp
• Text-to-speech using sherpa-onnx VITS
• Conversation memory across turns
• Barge-in support (interrupt while speaking)
• Follow-up listening without wake word
• Session recording for debugging
• Cross-platform audio player support (afplay, play, mpv, ffplay, aplay)