transcript_pipeline

automated: create gemini transcript: gemini_transcibe.py -> filename.json (same as mp3 filename) generate whisper transcript for json -> segments_with_whisper_and_gemini.json align the phrases using fuzzy logic -> segments_with_whisper_and_gemini_filled.json allign whisper and gemini transcripts whispx.py -> words_level_timesteps.json

obtain embeddings for gemini embeddings input is segments_with_whisper_and_gemini_filled.json perform clustering with 'Vaisesika Dasa' as the main speaker -> global_map.json asssociate speaker name with each segment within the transcript (single_merge_speaker.py) -> global_map.json, segments_with_whisper_and_gemini_filled.json -> filename.json

benchmark: run benchmark upload benchmark

manual: create splits and diarization: diariazation_from_segments.py