Empty TFLite output about tensorflowasr

[liuyibox]

Is this because the run_tflite_model.py file did not organize the input as the way the training dataset did? run_tflite_model.py directly feeds the raw audio into the tflite model instead of doing preprocessing, e.g., stft.

[liuyibox]

Is this because the run_tflite_model.py file did not organize the input as the way the training dataset did? run_tflite_model.py directly feeds the raw audio into the tflite model instead of doing preprocessing, e.g., stft.

I used the pretrained tflite models on the drive for inference and it does output the correct speech transcripts as expected. So this does not necessarily need preprocessing.

However, the weird thing is after I convert the h5 model on the driver to tflite and I inference on that tflite model, the inference outputs weird transcripts, how can this happen? Can you provide the h5 models for the 2 tflite models on google drive so that I can test if there are problems in the current "gen_tflite_model.py" file?

�s himsinengebe be theninsaidYgghimscamecamefacewrllblebleordlet barlet Ygg�Ybleil((bleblebleblebleblebe shoushouL

[liuyibox]

The "gen_tflite_model.py" works well. It turns out we definitely need more epochs to train the conformed models. After 20 epochs training, the conformer tflite model can outout "kutches me and i am bed" something meanful. I'm closing this issue.