09.12.2012 Views

The Kyma Language for Sound Design, Version 4.5

The Kyma Language for Sound Design, Version 4.5

The Kyma Language for Sound Design, Version 4.5

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

2. Select the first sample by clicking the Select button and using the file dialog to locate the sample<br />

Cmorph from the speech folder of the Samples folder of the Wavetables folder. <strong>Kyma</strong> will start<br />

playing the sample over and over with a 1 second silence between repetitions. To change the duration<br />

of the silence between repetitions of the sample to two seconds, type in a 2 in place of the 1, and<br />

press Enter.<br />

Listen to the sample and estimate the pitch range of the speech. <strong>The</strong>n click Next to continue.<br />

3. Choose the highest analysis frequency that is still lower than the lowest frequency you hear in the<br />

sample. In this example, the voice seems to go as low as 3 d or 3 c, so try 2 F as the setting.<br />

<strong>The</strong> analysis is done by a bank of band-pass filters equally spaced in frequency from 0 hz up to half of<br />

the sampling rate; the lower the frequency you choose here, the more filters there will be, and the<br />

closer those filters will be spaced in frequency. For 1 F, it uses 512 filters, <strong>for</strong> 2 F it uses 256 filters and<br />

so on (divide the number of filters in half each time you go up an octave in the analysis frequency).<br />

Press the Audition button to hear a real-time analysis and resynthesis with the analysis parameters as<br />

you have them currently set. Just below the Audition button, <strong>Kyma</strong> will print out the number of sine<br />

wave oscillators being used to resynthesize the sound from the real-time analysis. This number will<br />

vary depending on the analysis frequency you have chosen and the number of expansion cards in<br />

your Capybara.<br />

Try out a different analysis frequency by choosing 5 F and then clicking the Audition button again.<br />

Clearly this setting does not give you enough filters to adequately cover what is going on in the signal,<br />

so change it back to 2 F.<br />

4. In the last step, you effectively chose the number of bandpass filters to be used in the analysis. This<br />

next step controls the bandwidth of those filters. In the everpresent tradeoff between time and frequency,<br />

you can choose BestFreq and get the narrowest bandwidths (and the most time smearing), or<br />

you can choose BestTime to get the best time response (but the widest bandwidth and thus the worst<br />

frequency resolution), or you can choose any of the intermediate settings to get intermediate bandwidths<br />

and intermediate results with respect to time smearing and frequency resolution.<br />

Leaving the frequency set at 2 F, try out each of BestFreq, BestTime, BetterFreq, and BetterTime,<br />

clicking the Audition button after each choice to hear the differences. Generally speaking, you can<br />

tell when the bandwidth is too narrow if the result sounds like it has reverb or a delay on it, and you<br />

can tell when the bandwidth is too wide by a kind of roughness or distorted sound in the resynthesis.<br />

When you find the best sounding analysis (or the least-strange-sounding analysis if you are feeling<br />

negative today), click Next to continue.<br />

5. You are ready to try the first analysis! Click the button labeled Create Spectrum File. <strong>The</strong> name of the<br />

analysis file will be the name of the original sample followed by an s (<strong>for</strong> spectrum) and the number<br />

of analysis filters used (or, to put it another way, the number of analysis “tracks” that can be read by<br />

an OscillatorBank and resynthesized).<br />

Create your own wavetables folder called MyWavetables. In it, create a folder Spectra, and within<br />

Spectra create a folder called speech. Save the analysis file in this speech folder. <strong>The</strong> analysis tool<br />

automatically creates a SumOfSines <strong>Sound</strong> and places it into an untitled <strong>Sound</strong> file window <strong>for</strong> you.<br />

Just <strong>for</strong> fun, double-click on this <strong>Sound</strong>, change the Duration to On, and multiply the OnDuration<br />

by 10 to time-stretch the resynthesis.<br />

A straight spectrum is best <strong>for</strong> time-stretching or frequency-scaling, or when the two files you are<br />

morphing have no definite pitch to them. However, if the original samples have identifiable pitches<br />

in them, the morph will probably work better if you proceed to the next step: the quasi-harmonic<br />

analysis.<br />

If your original sample has an identifiable pitch to it, you should continue with the analysis by clicking<br />

on Set Quasi-Harmonic Parameters. Since CMorph (and indeed human speech in general) does<br />

have an identifiable pitch, click on that button now to continue.<br />

6. In this next section of the analysis, <strong>Kyma</strong> asks you to help identify the fundamental frequency of the<br />

original sample. <strong>The</strong> first step is to identify parts of the sound that do not have any identifiable pitch<br />

to them.<br />

151

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

Saved successfully!

Ooh no, something went wrong!