Garbage modeling


Adding words


Microphones


Open Mic


Real-time response


Sample rate conversion


Grammars, dictionaries, and "vocabularies"


Other FAQs


Other Sprex Products/Services


Exit>

This requires a bit of DSP hacking.  I've done and debugged this kind
of thing in the past in a not-very-clean way (simple linear
interpolation between samples) which works but sacrifices quality.
The theory of how to do it cleanly is in, for example, Bracewell's
Fourier Transform and its Applications, p194 (McGraw-Hill 1986),
basically you convolve the sampled output with the "sinc" function
(sin(pi*t)/pi*t) at each output time, presumably windowing the sinc
function values at some distance from zero.  Here's a non-debugged
pass at the core function.

signed short *double_sample_rate(signed short *in,int len,int order) {
  /* *in is an array of shorts of length len.  
   * order is the number of  sinc multiply-adds to do in the 
   * convolution that approximates each output sample.
   * free the returned buffer when you're done with it.
   * The return value is a pointer to the doubled-in-size output buffer
   * containing a doubled-in-sample-rate version of the input.
   */
  const float sinc[] = { 0.6366, -0.2122, 0.1273, -0.0909, 0.0707, -0.0579, 
		         0.0490, -0.0424, 0.0374, -0.0335, 0.0303, -0.0277 }
  int i,j,k,m;
  float sum;

  /* keep order between 1 and 12, the size of sinc[] */
  if (order > 12) order = 12; if (order < 1) order = 1;

  /* malloc output buffer */
  if (out = (signed short *)malloc(sizeof(signed short)*len*2)) {
    return(NULL);
  }

  /* foreach input sample generate two output samples */
  for (m = 0; m < len; m++) {
    out[m*2] = in[m];
    /* out[m*2+1] = sum over k = 0..order of sinc[k]*in[m+1+-(k+1)] */
    for (sum=0.0,k=0; k4) sample rate conversion you could for
example call double_sample_rate twice on a buffer of 12kHz data and
then extract every third sample to get 16kHz data.  Some faster
version could be made if you worked it out and avoided calculating the
3/4 of the samples that are irrelevant.

   From:
Message: