Suppose $x(t)$ is a signal containing a voice. How can one derive a signal playing the same voice faster.

I tried $x(t+\lfloor{100t\rfloor}/100)$ to double the speed; But it adds a buzz sound. Is there a better formula?

Clearly $x(2t)$ changes the voice. I want to change just the speed.

  • 1
    $\begingroup$ See other posts that mention “phase vocoder”; changes the speed without changing the pitch or vice versa $\endgroup$ – Dan Boschen Oct 28 '18 at 2:41

You can increase the sampling rate, instead of the signal $x(nT_s) = x(n/F_s)$, you can play it at say $\frac{11}{10}F_s, \frac{12}{10}F_s, \text{or}\ 2F_s$. So you have: \begin{align} x\left(\frac{n}{\frac{11}{10}F_s}\right) &=x\left(\frac{10n}{11F_s}\right) = x\left(n\frac{10}{11}T_s\right)\\ x\left(\frac{n}{\frac{12}{10}F_s}\right) &=x\left(\frac{10 n}{12F_s}\right) = x\left(n\frac{10}{12}T_s\right)\\ x\left(\frac{n}{2F_s}\right) &=x\left(\frac{1}{2}\frac{n}{F_s}\right) = x\left(n\frac{1}{2}T_s\right) \end{align} In MATLAB for instance, you can try the code below.

clear all

load handel.mat
sound(y, Fs) % original 
sound(y, 1.1*Fs) % at 1.1*Fs
sound(y, 1.2*Fs) % at 1.2*Fs
sound(y, 2*Fs) % at twice Fs

Gradually increasing the frequency to hear the changes in speed. You can try even higher, like 4 times, to hear the differences. The handel.mat file should be in there by default.


You $x(t)$ in your program is samples of the signal for a certain time duration, check how many samples you have per second (i.e. the sampling rate or frequency $F_s$), and $T_s = 1/F_s$. And $x(t) = x(nT_s) = x(n/F_s)$ with $n = 0, \ldots, N-1$ where $N$ is the number of samples. Taking a higher $F_s$ corresponds to taking smaller $T_s$.

| improve this answer | |
  • $\begingroup$ Thanks. But I have C# program which gets $x(t)$ as a continuous function and uses it to produce a wave file. I have the sampling machine code but I prefer to change $x(t)$ itself. Can you recommend a change in $x(t)$? $\endgroup$ – Minimus Heximus Oct 28 '18 at 12:09
  • 1
    $\begingroup$ @nano please see my edit. $\endgroup$ – Gilles Oct 28 '18 at 14:06
  • 1
    $\begingroup$ Isn't it the same as sampling $x(at)$ where $a$ is a constant? The problem with this is that the pitch changes. $\endgroup$ – Minimus Heximus Oct 28 '18 at 14:09
  • 1
    $\begingroup$ @nano search DSP.SE for pitch shift or associated techniques (like "time stretching"). This is a frequent question on this board and you can find some useful stuff in addition to the method that Gilles is suggesting to you here. $\endgroup$ – A_A Oct 29 '18 at 9:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.