Lab 4a – Analyze Your Rabiner-Sambur Implementation

 

There are several constants you had to set or compute in your implementation of the Rabiner-Sambur algorithm, e.g. ITU, IZCT, frame-size for smoothing. 

1.     List all these constants.

2.     Explain what each one means, what it is used for, and how you decided on the value.

3.     Run your program on the data files in /course/csu610/

In particular, run your program on the male and female words,

      two, three, four, six, nine

If your program is designed to look for more than one utterance in a file, try the two long files that have all the numbers.

4.     For each file, tell what your program did well and what did poorly.

5.     Look at the list you prepared in 1 and 2.  Are there constants you could change to make it work better on some of these speech files?  Does this make it worse on other files?

6.     Do back and look at your list of constants.

·       Which constants prevent your program from running in real-time?  (e.g. in immediate response to microphone input.

·       Which constants seem to be speaker dependent?

·       Which constants seem to be gender dependent?

·       Which constants depend on the sampling frequency?

·       Which constants depend on physical measurements, like seconds?

7.     How can you change your program to make it more generally applicable?

 

 

Lab 4b – Spectrograms

 

1.     Use the Matlab specgram function to make a narrowband spectrogram of the male and female numbers.wav files.

 

specgram(y,400,Fs,[],150) % for a narrow-band spectrogram *

·       From these two images, estimate F0 in the middle of the words: two, three, four, six, nine.

 

2.     Use the Matlab specgram function to make a wideband spectrogram of the male and female words: two, three, four, six, nine.

 

specgram(y,128,Fs,50,25); % for a wide band spectrogram *

·       From these ten images, estimate F0 in the middle of the words: two, three, four, six, nine.

 

(*  I got these parameters by experimenting and then asking a friend.  You might as well use them.)

 

3.     Did your numbers agree?  Which method was easiest?  Was gender a factor?