if I use an audio in which the range of the value is from -1 to 1, the vad is almost wrong,and the output data is also bad.
But if I offer a gain (at least 20dB+) to the input data, the vad works well and I just need to fix the output data.
So if I want to train a new model, how could I deal with the training data?
if I use an audio in which the range of the value is from -1 to 1, the vad is almost wrong,and the output data is also bad. But if I offer a gain (at least 20dB+) to the input data, the vad works well and I just need to fix the output data. So if I want to train a new model, how could I deal with the training data?