Feat/update whisper

[74th]

This pull request introduces several enhancements and refinements to the wakeword detection system using Whisper Server, focusing on improved configurability, more robust false positive avoidance, and clearer documentation. The most important changes include new environment variables for specifying wakeword keywords and ignore phrases, updates to buffer handling and detection logic, and expanded documentation with usage examples.

Wakeword Detection Configuration and Logic Improvements:

• Added support for configuring wakeword keywords and ignore phrases via new environment variables (STACKCHAN_WWD_WHISPER_SERVER_KEYWORDS and STACKCHAN_WWD_WHISPER_SERVER_IGNORE_PHRASES), allowing for flexible keyword and false positive management. [1] [2]
• Introduced a minimum buffer duration (min_buffer_seconds) before running inference, reducing premature or spurious detections. [1] [2] [3]
• Updated the logic to ignore transcripts containing any configured ignore phrases instead of just the prompt, further reducing false positives from Whisper model outputs. [1] [2] [3]

Documentation Updates:

• Expanded documentation in docs/server_ja.md and docs/websocket_protocols_ja.md to explain the new environment variables, their intended usage, and provided practical configuration examples for wakeword detection. [1] [2]

Codebase and Usability Enhancements:

• Improved buffer and PCM handling by adding helper methods for byte/second conversion and buffer truncation, and clarified environment variable prefixes for consistency. [1] [2]
• Minor script adjustment in run-whisper-server-small.sh to add the --convert option for better compatibility.