Speakr is a personal, self-hosted web application designed for transcribing audio recordings
Python
Fix clear-voice-embeddings: null the real average_embedding column, not a phantom attribute
Murtaza Nasir10b12beeff51
diff --git a/src/api/speakers.py b/src/api/speakers.pyindex 2151edf6..4224b069 100644--- a/src/api/speakers.py+++ b/src/api/speakers.py@@ -388,8 +388,14 @@ def clear_speaker_embeddings(speaker_id):if not speaker:return jsonify({'error': 'Speaker not found'}), 404- # Clear all embeddings- speaker.voice_embeddings = None+ # Clear all embeddings. NOTE: the matchable vector lives in the+ # `average_embedding` column (and the history in `embeddings_history`)+ # — there is NO `voice_embeddings` column. The previous code set a+ # phantom `voice_embeddings` attribute that never persisted, so the+ # real average_embedding survived and voice matching kept working+ # after a "clear". Null out the actual columns.+ speaker.average_embedding = None+ speaker.embeddings_history = Nonespeaker.embedding_count = 0speaker.confidence_score = None