Due the results of the experiment for establishing the threshold in silence remover, we got to the conclusion that the best idea is to amplify some signals and establish this threshold at 300.
The problem that appeared is when to amplify? At this moment there are problems in detection of maximum amplitude of a wav file. The changing of amplitude program is working, but it is not known the factor.
I tried to read the file as a byte[] and get maximum value in absolute value - but it is always 128, no matter the file (tested on 10 files).
I tried to get max, min and avg of the nAverage on a frame, and avg/frame, avg/file and sum of the byte[]. The following results were obtained:
/ nAverage / Elements
File Max(avg) Min(avg) Avg(avg) AVG/frame SUM AVG/File rms
bernd 1 3424 0 683 4844 683051 10 5895
bernd 2 3420 0 583 1099 196894 2 7890
bernd 3 3633 54 529 5093 911767 11 7159
bernd 4 2863 76 771 75 11561 0 8053
marina A 1 3294 0 439 -995 -216997 -2 9690 (There were problems at silence removing!)
marina A 1/A 3597 0 640 -1161 -253232 -2 9604 (File amplified with the program with 1.6
640/439 = 1.46
Observation: seems that most important is Avg(avg)!!
OTHER EXPERIMENT:
We used the normalization program made by dennis and the Silence removal program.
The values for the parameters (normalizing parameter, silence removal parameter)were: (0.6, 300) and (1,500)
In both cases not all wav files were ok, but from listening and seeing the files with praat, the first combination is better.
The following problems were found:
- in some of the files the noise was amplifed and because of this not removed in the removal operation (ricardo2.wav)
- in some of the files some letters were cut
- after removing silence, the sound is not normalized anymore
--
GavrilaMonica --
17 May 2004