Results from encoding with 44kHz model at 8kbps does not yield an 8kbps encoding

[fred5825]

Hello,

I am running the example script with the 44kHz model at a 8kbps bitrate but when I manually calculate the bitrate it ends up being 16kbps. I have encountered the same issue with running the 16kHz model at 8kbps bitrate, I calculate the actual bitrate to be ~20kbps. The audio sample I used for both are sample at 44.1kHz and 16kHz respectively and both mono channels. Curios if anyone had gotten this issue?

[liangjindeamo-yuer]

I have the same problem and the results are the same as yours! The actual bitrate is ~20kbps, what's wrong with it?

[listener17]

@fred5825 "44kHz model at a 8kbps bitrate but when I manually calculate the bitrate it ends up being 16kbps. " --- maybe you're using the 16 kbps model?

[KemalAltwlkany]

Hi, I would like to report unexpected results as well in terms of bitrate.

My colleague and I have used the examples provided from the main github page. We applied the compress method to a 9.62 second audio file in .wav format. The resulting bitrate is:

• ~16kbps for the 44.1kHz model
• ~57kbps for the 24kHz model.

Generally the library is poorly documented with not enough explanation as to what the values are which are returned from the encode method (z, codes and latents) as it is unclear which should be used for storing compressed audio. We inspected the source code and concluded that the compress method stores the variable codes as uint16 into a DACFile. When using method decompress, the codes variable is used to recreate z from which audio is reconstructed.