GitLab

Hi Everyone, I'm circling back to an element I requested help with a month or so ago regarding the export of a network to protobuf format. You all were very helpful in providing me the input shape parameters for the data at the time, but I am now needing to use the export function for a different set of files and a bit turned around as to how to determine the the width and height values. I am using the following spectrogram parameters -

{
    "spectrogram": {
	"duration": "10.0 s",
	"rate": "8000 Hz", 
        "window": "0.128 s",
        "step": "0.064 s",
        "freq_min": "500 Hz",
        "freq_max": "4000 Hz",
        "window_func": "hamming",
	"normalize_wav": "true",
        "type": "MagSpectrogram",
	"transforms": [
            {"name":"reduce_tonal_noise"},
            {"name":"normalize", "mean":0.0, "std":1.0}
        ]
    }
}

My questions are:

1. How do I translate this into the appropriate height and width values for the input_shape argument?
2. The documentation for this argument states the "The number of instances and number of channels are commonly 1" - I understand the channel being 1 but what is the "number of instances" referring to and are there times when it is not 1?

Thank you so much in advance!