cc [ flag... ] file... –lmlib [ library... ] #include <mlib.h> mlib_status mlib_SignalMelCepstral_S16(mlib_s16 *cepst, mlib_s32 cscale, const mlib_s16 *signal, void *state);
The mlib_SignalMelCepstral_S16() function performs cepstral analysis in mel frequency scale. The user supplied scaling factor will be used and the output will be saturated if necessary.
The first two steps of mel scale cepstral analysis is the same as in general cepstral anaysis. After the logarithm of the spectrum magnitude is obtained, it is converted into mel frequency scale before the inverse Fourier transform.
+-----------+ +-----------+ | Linear | | Inverse | ... ------>| to |------->| Fourier |-----> X'(k) | Mel Scale | X''(m) | Transform | c(n) +-----------+ +-----------+
where X'(k) is defined in linear frequency scale and X''(m) is defined in mel frequency scale.
The mel frequency scale is defined as following.
freq_mel = melmul * LOG10(1 + freq_linear / meldiv)
where freq_mel is the frequency in mel scale, freq_linear is the frequency in linear scale, melmul is the multiplying factor, muldiv is the dividing factor.
Optionally, a bank of band pass filters in linear frequency scale can be used below the bank of band pass filters in mel frequency scale, as shown below in linear frequency scale.
0 f1 f2 f3 fp fp+1 fp+2 fp+3 fp+q |---|---|---| ... |---|----|-----| ... | ... -> freq
where fp = melbgn, fp+q = melend, p = nlinear, q = nmel; the filters number 1 to p are defined in linear frequency scale which have equal bandwidth in linear frequency scale; the filters number p+1 to p+q are defined in mel frequency scale which have equal bandwidth in mel frequency scale and increasing bandwidth in linear frequency scale.
See Digital Signal Processing by Alan V. Oppenheim and Ronald W. Schafer, Prentice Hall, 1974.
See Fundamentals of Speech Recognition by Lawrence Rabiner and Biing-Hwang Juang, Prentice Hall, 1993.
The function takes the following arguments:
The cepstral coefficients.
The scaling factor of cepstral coefficients, where actual_data = output_data * 2**(-scaling_factor).
The input signal vector, the signal samples are in Q15 format.
Pointer to the internal state structure.
The function returns MLIB_SUCCESS if successful. Otherwise it returns MLIB_FAILURE.
See attributes(5) for descriptions of the following attributes: