search

qiuqiangkong / audioset_tagging_cnn

MIT License
1.35k stars 255 forks source link

First float32_to_int16, and then int16_to_float32? #37

Closed sysuzyq closed 3 years ago

sysuzyq commented 3 years ago

during training, you transform the waveform from float32 to int16, and then back to float32. could you tell me why ?

but in pytorch/inference.py , you don't do this. could you tell me why ?

RicherMans commented 3 years ago

He does that to save some space. The raw data is ~ 1.1Tb, and by just this simple transformation, the datasize gets halved without much of a drawback.

sysuzyq commented 3 years ago

He does that to save some space. The raw data is ~ 1.1Tb, and by just this simple transformation, the datasize gets halved without much of a drawback.

thanks