What's the usage of n_frame in the .tsv file?

[RuiRuihigh]

HI. Could you tell me the usage of the n_frame in the .tsv file and how to get the number of this parameter for my own audio if I want to add some new audio? I modified the number to some random number, but the model can still work.

[XiaoshanHsj]

actually, it is not used. During experiments, we use "frames" to filter out some audio, whose duration is longer than 60s

[RuiRuihigh]

So, does that mean the model cannot completely receive an input audio longer than 60s? If it is, can I modify this limitation?

[XiaoshanHsj]

We did not use such long audio clips to train WavLLM. However, I think it can inference these audios, please have a try on 60s audio clip to do the ASR task to check.