Speech Processing System

Applications of Programmable DSP Devices A Speech Processing System • Processing depends on nature of application o e.g. filtering to improve speech quality • Here pitch period estimation • Fundamental frequency estimation • Application o Speech identification o Speech Verification o Aids to handicapped How speech is generated? • Vocal chord influences the voice quality • and fundamental frequency

Autocorrelation to estimate fundamental frequency

Computation:

n − k −1

Rn (k ) = ∑ y (n + m) y (n + m + k ) m =0

can have three different values: y(n+m)y(n+m+k)=0 if y(n+m)=0 or y(n+m+k)=0 =+1 if y(n+m)=y(n+m+k) =-1 if y(n+m) ≠ y(n+m+k) •

Process Steps: o Filtered with 900 Hz low pass analog filter o Sampled at the rate of 10 KHz o Signal samples are segmented or sectioned o 30 msec duration o Interval 10 ms o Overlap 20 ms o Threshold level o Above threshold Æpresence of speech o Below threshold Æ silence

Implementation on the TMS320C54xx Processor: • Speech samples were recorded in voice recorder tool in Windows 98 • Autocorrelation the most computation intensive part • Implemented in DSP • Advantage o Circular buffer for real time signal o MAC o Zero overhead looping