[Step 3](https://github.com/adhadse/Deepdub/blob/master/3_Clustering_audio_segments_for_speaker_diarization.ipynb) Analyze cosine similarity score (using heatmap) of same speaker's embeddings to learn about the generated embedding's usefulness. You can take inspiration from and even use [Resemblyzer](https://github.com/resemble-ai/Resemblyzer). 