Implemented OpenAI STT spec compliance tests: ## Changes Made 1. **routes.py** (src/librechat_audio/routes.py): - Added timestamp_granularities parameter supporting segment and word values - Implemented _build_segments() to generate OpenAI-spec segments with timestamps - Implemented _build_words() to generate word-level timestamps - Implemented _build_srt() and _build_vtt() for proper subtitle formats - Added _calculate_duration() to calculate audio duration from transcription result - Updated verbose_json response to include segments/words based on granularity 2. **test_transcriptions.py** (tests/test_transcriptions.py): - Added TestTimestampGranularities class with 4 new tests - Added TestVerboseJsonSpec class with 6 OpenAI spec compliance tests - Added TestSRTVTTFormats class with 3 subtitle format tests - Created mock dataclasses for testing ## Test Results All 27 transcription tests pass, 104 total tests pass. ## Note Unable to push to remote due to SSH key issues. Changes are committed locally on branch LCA-14.