Hello, I was curious if there was a way to increase or decrease the percentage overlap between frames when calculating MFCCs for an audio file using librosa.feature.mfcc. After reading through the documentation for a while I do not see a way to do this but thought I would create an issue to ask. Currently implementing a speaker recognition program and wanted to play around with the overlap percentage to see if it could improve my results. Thanks in advance.
Overlap is controlled by the hop_length parameter in all spectrogram-like functions. The default frame length is 2048 (for STFT operations), and the default hop is 512, which results in 75% overlap.
Great, thanks! That helps a whole lot.
25ms 10ms, 2048 512,,,those how to build relationships
Most helpful comment
Great, thanks! That helps a whole lot.