Distinct default sampling rate in CLAP(version='2023')

[kamilakesbi]

The CLAP model version 2023 process audios sampled at 44100 Hz according to this configuration file).

However, when we initialise the model, the htsat module still uses this other config file to define some of its parameters, as we can see here. In this config file, the sampling rate is set to 32000. As a result, the LogmelFilterBank is initialised with a sampling rate of 32000.

Is this behaviour expected ?

Thanks!

[Blinorot]

Hello, @bmartin1

Could you, please, have a look at this issue? It is not only about Sampling Rate, but also about Fmax. The values used in the code (from config.py here) are different from those in yml model config and the original 2023 version paper

Thank you!