Sound Synthesis II: SMS Frequency/Time Transformation Examples

Changing the pitch without changing the duration
Changing duration without changing pitch
Changing Even/Odd harmonics
Exercises

Here is a clarinet sound used for the demos on this page: cl-mf-c4.wav. To make good frequency/time transformations with SMS, you need to successfully segregate the deterministic components of the sound from the stocastic components; Otherwise, you will get noticeable artifacts appearing in the transformed sound (which you may or may not want).

To test that you have separated the noise from the pitched part of a sound, try to remove all of the pitched part from the sound and listen to the noise. Here is an example of how to do that:

Analysis file:

InputSoundFile cl-mf-c4.wav OutputSmsFile temp.sms nSines 60 HighestPitch 269 SineModel 1 FrameRate 344.53125 LowestPitch 254

Synthesis file:

InputSmsFile temp.sms OutputSoundFile 1.wav Type 2 SamplingRate 44100

The above analysis shows that the attack and release are not quite right. It is good enough for now, but you could try modifying the parameters AttackReanalysis for better results, as well as MagThreshold. Notice that Type = 2 in the synthesis file means to synthesize the residual sound (usually noise) only. The FrameRate of 344.53125 is needed to make a window size of 128 samples at 44100 Hz.

Changing the pitch without changing the duration

The following example analysis/synthesis score-file pair will generate a sound which is an octave higher than the original:

The output sound, 2.wav, does not contain any of the residual sound, because the Type was set to 1 which means synthesize only the sinusoidal components from the analysis.

To do non-octave transpositions, you can refer to the Pitch Tranpositions Scalings page in order to calculate the correct value for the FreqSin synthesis parameter.

Changing duration without changing pitch

The SMS synthesis parameter TimeStretch can be used to change the duration of the sound independently from the pitch. The following example analysis/synthesis pair will create an output sound which is 4 times longer than the original sound.

Refer to the NoUnvoicedTimeStretching parameter if you are changing the duration of a speech sound which you want to sound more natural.

Changing even/odd harmonics

There are many amplitude modification parameters in the SMS analysis parameters. Below is an example of removing the even and odd harmonics from a clarinet sound. For some reason the AmpSineOdd parameter controls the amplitude of the even partials, while the AmpSineEven parameter controls the amplitude of the odd partials.

This example removes the odd harmonics:

This example removes the even harmonics:

Stretched/Shifted harmonics

The SMS synthesis parameter FrequencySineStrech can be used to stretch the harmonics, making them more or less harmonic. For example, if the harmonics were 100, 200, 300; you could stretch them like this 100, 205, 310, etc. Here is an example with a stretched set of harmonics:

Analysis file:

InputSoundFile cl-mf-c4.wav OutputSmsFile temp.sms nSines 16 HighestPitch 269 SineModel 1 FrameRate 344.53125 LowestPitch 254

Synthesis file:

InputSmsFile temp.sms OutputSoundFile 6.wav Type 1 FreqSineStretch 0.05 SamplingRate 44100

The above example has a frame-rate problem caused by the sms program which will have to be identified later.

The shift operation keeps a constant interval between harmonics. The FreqSineStretch parameter keeps the frequency difference between harmonics constant, but not the interval. Here is an example of shifting the harmonics by 10 Hz:

Exercises

Choose some sound(s) to transform in frequency and/or pitch. Do several of the example transformations on the sounds and present them to the class on Thursday.
Refer to the Analysis Parameters and Synthesis Parameters as you create your example transformations.
Try some transformations on speech. For example, try NoUnvoicedTimeStretching on a voice which is slowed down by a factor of 3.
Give a speech sound a clarinet-like timbre.
Extra Credit: Transform the pitches of several instrument either from very high to very low, or vice-versa, keeping the duration constant. Have other people listen to the transformed pitch and see how well they can identify the instrument with/without the transpositions.