 |
|
 |
|
|
| ||||
| ||||
|
| ||||
 |
 |
| |
| |
 |
|
|
|
|
|
|
|
|
|
|
| |
This effect was by far the most difficult of them all to derive. Which makes sense since this was the effect that spawned the project in the first place. Before I discovered a good method to get that mechanized robot sound I tried many different methods. Originally I was focused on taking two signals, one speech and one melodic, running them both through the encode process and then modulating different parts of those signals during the interpolation process. I tried just about every variety of modulations to every part of the signal. I got some nice distortion sounds but NO robotic sound. I then dropped the idea for a while and got back to producing the other effects and finishing the FT phase vocoder. As suggested by Professor Ellis, I began focusing on experimentation with the phase itself and retired the modulation idea. During this experimentation I also began playing with the Hann Window parameters and overlap again. This led to the combination of parameter changes that in the end created an aesthetic synthetic voice sound.
So how is this sound created? In the end the answer was very simple, as most good solutions are. To achieve a quality robotic sound all that is needed is to remove the phase advance and lower the hamming window (and FFT) point size. Reducing either of these alone does not do much. It only when they are done together does a rich effect come about. When a signal is ran though the phase vocoder and the Hann Window is reduced (and the timescale is 1:1), you do not notice much difference. If however you do a time stretch you will notice distortions creaping in. The distortion introduced with the reduced Hann Window size and the static phase effect of eliminating the phase advance together create a great robot sound.
Spectrogram of wave file 1 processed with robot.m (with a 256 point Hann Window)
You can see from the spectrogram that the robot effect has altered the wave. You can see that the effected signal has its content smoothed and smeared slightly. This probably indicates a less richness in the frequency spectrum and reduced granularity. Also there are lines running through the graph. This is most likely due to the staleness of the phase since it is not being properly adjusted with the phase advance. The static phase helps to create the robot sound by distorting the fequency spectrum as seen in the graph above. Those lines probably occur near bin edges since the phase advance effects those frequencies the most.
Wave File Examples:
Wave File 1 - 512 point Hann WIndow
Wave File 1 - 256 point Hann Window
Wave File 2 - 256 point Hann Window