List view
v2 goals: - Significantly improve Speech to Text performance - Eliminate external dependencies and make SkyEye as fully a self-contained program as possible Technical work items: - Improve STT performance by replacing OpenAI Whisper with Parakeet V3 - Ideally, eliminate build dependency on whisper.cpp - Ideally, eliminate runtime dependency on external whisper model files, instead embed Parakeet within SkyEye - Ideally, simplify codebase by removing cloud speech recognition in favor of fast local STT - Remove TTS subprocessing by replacing embedded Piper and macOS speech synthesis subprocess with Pocket TTS - Eliminate fragile dependency on embedded piper binary continuing to work - Simplify installation on macOS - Eliminate risk of Siri voice and pronunciation significantly changing in future macOS updates - Minor cleanup of public APIs Non-goals: - Significant changes to player experience other than STT/TTS changes - No new commands, workflows, etc. - Refactor of messy controller and radar packages - Use of SOTA LLMs/AIs for controller logic (maybe someday when these models can run locally and affordably)
No due date•1/3 issues closed