Speech signal processing - concept (II): amplitude spectrum (STFT spectrum), Mel spectrum [the deep learning of speech mainly uses amplitude spectrum and Mel spectrum] [extracted with librosa or torch

One 、 What kind of spectrum is used in the deep learning of speech ？
answer ： With amp spec and melspec Mainly , Usually it can be used librosa Library or torchaudio Library for extraction .

Two 、 that Fbank、MFCC Well ？
answer ： Because it contains less information , It is no longer suitable for this big data era . But some tasks, because of their special nature , Still use MFCC Spectrum . Such as emotional voice conversion task .