Fun Project
Take 1: Mathieu Amalric with a similarity of 94.53%
Take 2: John Turturro with a similarity of 94.24%
Take 3: Garry Marshall with a similarity of 93.76%
Curious to know how this can be improved to always return a single name?
Thank you for your feedback! The reason you’re seeing multiple similar results is that our current system relies on a simplified model with a limited dataset and single voice averages (centroids) for efficiency. This approach means that voices from certain celebrities are clustered closely together in the model. If you use the same audio file, the results will remain consistent because the process is deterministic.One potential improvement would be to use multiple centroids per speaker, which could capture a broader range of voice variations and better distinguish between similar voices. This enhancement could lead to more definitive matches in the future.