On the IMXRT1060, I have run the example tensorflow_lite_kws. This example use frame size of 40 ms, hop 20 ms to split the audio, then do FFT, and turn into MFCC. While the Yamnet use the 0.96s length, 0.48s hop to split the audio. Then, each frame do the STFT with window 25ms, hop 10ms, the different with available example. The question is, How can I implement the Yamnet on this microcontroller, can you suggest necessary steps to do it? I find it difficult because with tensorflow_lite_kws, each frame do 1 FFT; while Yamnet, each frame do 96 STFT. Thanks in advance.
The input of tensorflow_lite on IMXRT1060 is always uint8_t? Can we change this input into int16 or float? I ask this question because I see this convert:
