Presentation is loading. Please wait.

Presentation is loading. Please wait.

Topic: Pitch Extraction

Similar presentations


Presentation on theme: "Topic: Pitch Extraction"— Presentation transcript:

1 Topic: Pitch Extraction
Kishore Prahallad Carnegie Mellon University & International Institute of Information Technology Hyderabad Speech Technology - Kishore Prahallad

2 Objective of this Lecture
Describe the extraction of pitch using autocorrelation function Speech Technology - Kishore Prahallad

3 Speech Technology - Kishore Prahallad (skishore@cs.cmu.edu)
Pitch Period Time Taken to complete one cycle of vibration of vocal folds Pitch Period is also referred to as fundamental frequency or F0 Measured as time difference between two major peaks in the voiced speech signal Pitch is observed only in voiced regions Speech Technology - Kishore Prahallad

4 Speech Technology - Kishore Prahallad (skishore@cs.cmu.edu)
Pitch Marks Pitch extraction is done through autocorrelation based algorithm Implementation details may be necessary to tune the pitch Tune the parameters of pitch extraction to tune to the specific speaker (your voice talent) Speech Technology - Kishore Prahallad

5 Algorithm: Autocorrelation based Pitch Extraction
1. Filter the speech signal Pitch range is 40 – 400 Hz ( Hz for male, and Hz for female) Use a low pass filter - restore the frequency components less than 800 Hz Use a high pass filter – restore the frequency components greater than 40 Hz Speech Technology - Kishore Prahallad

6 Speech Technology - Kishore Prahallad (skishore@cs.cmu.edu)
Algorithm.. 2. Divide the signal into shorter analysis window 3. For each short analysis window Take autocorrelation of the signal Speech Technology - Kishore Prahallad

7 Autocorrelation Function r[k]
Speech Technology - Kishore Prahallad

8 Autocorrelation of a Vowel signal
Speech Technology - Kishore Prahallad

9 Autocorrelation of a Vowel signal
Remember - Pitch is the time difference between two major peaks. Here first peak is at 0, and the second peak is at 110. Hence pitch is 110 samples Pitch in Sec (t) = 110/16000, – sampling freq Pitch in Hz = 1 / t = 16000/110 = 145 Hz Pitch = 145 Hz implies this is a male speech r[0] r[110] Autocorrelation Speech Technology - Kishore Prahallad

10 How do you know r[110] is the second major peak?
To know that r[110] is closer to r[0], we use peak picking algorithm on the autocorrelation function. Peak picking algorithm search for a peak in a *specified* region. This region is the tunable parameter 40 – 200 Hz Range: r[400] – r[80] for male The numbers in r[] are calculated for Hz sampling frequency 200 – 400 Hz Range: r[80] – r[40] for female Speech Technology - Kishore Prahallad

11 Speech Technology - Kishore Prahallad (skishore@cs.cmu.edu)
Algorithm contd… 2. Divide the signal into shorter analysis window 3. For each short analysis window 3.1 Take autocorrelation of the signal 3.2 Pick the second major peak and obtain pitch information in sec. Note: This is a simple yet a good pitch extraction algorithms. (We use this method for all our discussion) There are *many* ways of extracting pitch from the speech signal Ex: FFT based, linear prediction based etc. Speech Technology - Kishore Prahallad


Download ppt "Topic: Pitch Extraction"

Similar presentations


Ads by Google