Unformatted text preview: UCLA Dept. of Electrical Engineering EE214A: Digital Speech Processing Problem Set 3 Due: 1/31/2011 Reading Assignment: Chapters 6 and 7 (till Section 7.5).
1. The con guration shown in the gure below is an idealized vocal tract during the production of the vowel /u/. The dimensions are as follows: lb = 3 cm lc = lm = 2 cm Ab = A2 = 3 cm2 Ac = Am = :3 cm2 (a) Write expressions to calculate the rst four formant frequencies for a male
vocal tract (total length=17 cm). Indicate which tube(s) give rise to each formant. Assume that the speed of sound in air is c = 35000cm=sec and that the tubes can be decoupled at high frequencies. the female formants represent a scaled version of the male formants? If so, what is the scaling factor? female, 200 Hz, how would these values a ect your calculations. Explain.
b (b) Repeat part(a) for a female vocal tract with a total length of 14 cm. Do (c) If you knew that the pitch (F0) for the male speaker is 100 Hz and the
Ac A2 Am 1 2. Problem 6.1. 3. Problem 6.9. 4. Problem 6.17. 5. Problem 7.5. 6. Problem 7.6. 7. Problem 7.12. 8. Wavesurfer Question: Download the software wavesurfer and the provided sentences. Open the le s5.wav in wavesurfer, and choose the speech analysis option. You will see a graphic comprised of three sections. The top part is the speech waveform, the middle section is the spectrogram (including formant tracking), and the bottom part is the pitch tracking of the speech. Note that you are able to scroll through the sentence using the scroll bar at the very bottom. Right click on the spectrogram. Choose the properties option. Now choose the spectrogram tab. The settings should be set to the default settings. The default window used for the spectrogram analysis is a Hamming window of 64 points (4 ms). Change the window length to 400 points, and press OK. Describe the di erences in the spectrogram using the 64-point window versus using the 400-point window. Which window would you expect to be better for formant detection? Why? Which window would you expect to be better for pitch detection? Why? 2 ...
View Full Document