22.09.2015 Views

of Microprocessors

Musical-Applications-of-Microprocessors-2ed-Chamberlin-H-1987

Musical-Applications-of-Microprocessors-2ed-Chamberlin-H-1987

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

582 MUSICAL ApPLICATIONS OF MICROPROCESSORS<br />

waveforms, but blind application <strong>of</strong> it can lead to errors with rapidly changing<br />

waveforms and unpitched sounds.<br />

In truth, human hearing probably utilizes both kinds <strong>of</strong> information. It<br />

is unlikely that the pitch <strong>of</strong> high-pass (I-kHz) filtered 60-Hz buzz is determined<br />

from a frequency analysis because <strong>of</strong> the resolution necessary to account<br />

for the very precise judgments people are capable <strong>of</strong>. On the other<br />

hand, waveforms are easily contrived in which the periodicity is not easily<br />

spotted visually yet they are also accurately judged. Then we have the case <strong>of</strong><br />

bells and chimes, which are decidedly nonperiodic and therefore nonharmonic<br />

yet produce a strong sensation <strong>of</strong> pitch. Last, but not least, the<br />

existence <strong>of</strong> a tone that sounds as if its pitch is continuously risingforever<br />

I-underlines the fact that human pitch perception is a very complex<br />

topic indeed.<br />

Time-Domain Methods<br />

The simpLest pitch detection methods work in the time domain, where<br />

the waveform itself is examined for periodicity. In fact, it is the period that is<br />

measured that requires a division to determine the corresponding frequency.<br />

Most time-domain methods were developed for analog implementation;<br />

however, they can be readily implemented as a program for processing sampled<br />

data. One problem with sampled data is that the period determination<br />

is normally made in terms <strong>of</strong> an integer number <strong>of</strong> samples. This can lead to<br />

significant errors unless the sample rate is high or interpoLation is used, both<br />

<strong>of</strong> which increase processing time.<br />

In order to improve pitch detector performance, it is common practice<br />

topreprocess the signaL. Typically, this consists <strong>of</strong> low-pass filtering to remove<br />

high-frequency harmonics and noise that may contribute to jitter or confuse<br />

the detection algorithm. An obvious question, then, is what the cut<strong>of</strong>f<br />

frequency should be. If the sounds being analyzed have a strong (but not<br />

necessarily dominant) fundamental, then it is appropriate to set the cut<strong>of</strong>f<br />

just above the highest fundamental expected. Often, dynamic filters are<br />

utilized, in which case the cut<strong>of</strong>f can track just above the current fundamental<br />

frequency. If the fundamental is dominant, as it <strong>of</strong>ten is in direct string<br />

pickups and mouthpiece microphones, such preprocessing may be sufficient<br />

to allow a simple zero-crossing detector to do the actual period detection.<br />

Even if the fundamental is not dominant, repeated integration can make it<br />

dominant. U nforrunately, the integrators emphasize low-frequency transients<br />

and may actually blank out the zero-crossing detector for hundreds <strong>of</strong><br />

'This is usually demonstrated as a continuous upward sweep or an ascending musical<br />

scale that never stops. [n reality, the effect is much like the stripes on a barber pole,<br />

and, in fact, a spectrogram <strong>of</strong> the former example is an endless series <strong>of</strong> upwatd<br />

sloping diagonal bars. For a scale, the continuous rise is simply quantized at musical<br />

scale pitches.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!