coded two types of confidence: 1. input_confidence = the number of matched hashes / the number of generated hashes for the query window. 2. fingerprinted_confidence = the number of matched hashes / the number of predicted song’s hashes fetched from DB -- coded confidence distribution for each type -- divide each type into two parts: correct prediction confidence and fault prediction confidence. (the same for the confidence distributions) — run the analysis for a 5-sec query window

- half sec overlapping left side.

- overlapping length 1 sec on the left side.

this report shows that the query time is not fixed for different runs of the same exp. the same exp can give two different query times if runs twice.

- order of table rows are: 2-sec, 6-sec, 5-sec, 3-sec, 4-sec, full window - frequency 12000 - uniform splitting

- order of table rows are: full window, 5-sec, 4-sec, 2-sec, 3-sec - frequency 12000 - no splitting

- order of table rows: 6-sec, 5-sec, 4-sec, 3-sec, 7-sec, full window - no splitting - frequency 11500

- order of table rows are: 3-sec, 4-sec, 5-sec, 6-sec, 7-sec, full window - frequency 11500 - no splitting

fingerprinting with frequency 11500 (this will change to 12000 after using pydub) inference with frequency 12000 (which is produced by loading and save 11500 audios with torchaudio) - 4-sec window - no splitting

use 12000 frequency for both fingerprinting and inference - 4-sec window - no splitting

I created fingerprints for one long audio and made inferences on 9 different audios and this long audio. the long audio I used for creating fingerprints: - 34 min: chunk_Khalid Al-Jileel_Khalid Al-Jileel Mushaf_012_0