I have a 4 class classification problem using speech data. The speech signals are of variable length ranging from 1 sec to 8 sec. I am making chunks of 3 secs by chopping the longer utterances and zero padding the small utterances.
Now, when testing, the final label for an utterance should be aggregate of labels of the smaller chunks produced from that utterance. What is the recommended way of doing it ?
No of chunks per utterance also varies- eg-
utterance 1 - 3 chunks of 3 secs each
utterance 2 - 2 chunks of 3 secs each.
for utterance 2, suppose if label comes to be class 1 for chunk 1 and for chunk 2 , it comes class 4. How to decide which label to assign to the full utterance?