Why does transcription have a significant drop in quality while running over docker?

[Deuos]

While using the small english model version 0.15, I noticed that when downloading it and running locally the transcription is more accurate, than using WebSockets to pass the audio and running it through the docker container. Why does this happen? Is it because of websockets or the image itself?

[nshmyrev]

It is hard to guess, actually results must be identical

[Deuos]

Ok, redid tests using the official test16k.wav, the results were identical, the issue may have been with my audio examples for some reason. One question, the sample rate must be 16000 right? And the bit rate doesn't matter correct?

[nshmyrev]

Sample rate is usually 16000 yes, bitrate for 16bit per sample is 256kbps

[Deuos]

Ok thank you, that examples it.