Add filtered multi-speaker parquet with WAV bytes and speaker_id

initial commit