mglsp2sp

Functions

int main(int argc, char *argv[])

mglsp2sp [ option ] [ infile ]

• -m int

  • order of line spectral pairs $(0\le M)$

• -a double

  • all-pass constant $(|\alpha |<1)$

• -g double

  • gamma $(-1.0\le \gamma <0)$

• -c int

  • gamma $\gamma =-1/C$ $(1\le C)$

• -l int

  • FFT length $(1\le L)$

• -s double

  • sampling rate $(0<F_s)$

• -k int

  • input gain type

    • 0 linear gain

    • 1 log gain

    • 2 without gain

• -q int

  • input format

    • 0 frequency in rad

    • 1 frequency in $\pi$ rad

    • 2 frequency in kHz

    • 3 frequency in Hz

• -o int

  • output format

    • 0 $20{\log }_{10}|H(z)|$

    • 1 $\log |H(z)|$

    • 2 $|H(z)|$

    • 3 $|H(z){|}^2$

• infile str

  • double-type mel-LSP

• stdout

  • double-type spectrum

Parameters:

• argc – [in] Number of arguments.

• argv – [in] Argument vector.

Returns:

0 on success, 1 on failure.

See also

lpc2lsp

class MelGeneralizedLineSpectralPairsToSpectrum

Convert mel-LSP to spectrum.

The input is the $M$-th order line spectral paris:

$$\begin{array}{cccc}K,& \omega (1),& \dots ,& \omega (M),\end{array}$$

where $K$ is the gain. The output is the $(L/2+1)$-length log amplitude spectrum:

$$\begin{array}{cccc}\log |H(0)|,& \log |H(1)|,& \dots ,& \log |H(L/2)|,\end{array}$$

where $L$ is the FFT length.

The transfer function of the all-pole filter is given by

$$H(z)=K\cdot A^{1/\gamma }(z).$$

Thus,

$$\log |H(z)|=\log K+\frac{1}{2\gamma }\log |A(z){|}^2.$$

If $M$ is even,

$$\begin{array}{r}\begin{array}{rcl}|A(\omega ){|}^2& =& 2^M\{{\cos }^2\left(\frac{\tilde{\omega }}{2}\right)\prod _{m=1,3,\dots }^M(\cos \tilde{\omega }-\cos \omega (m){)}^2\\ & & +{\sin }^2\left(\frac{\tilde{\omega }}{2}\right)\prod _{m=2,4,\dots }^M(\cos \tilde{\omega }-\cos \omega (m){)}^2\},\end{array}\end{array}$$

else

$$\begin{array}{r}\begin{array}{rcl}|A(\omega ){|}^2& =& 2^{M-1}\{\prod _{m=1,3,\dots }^M(\cos \tilde{\omega }-\cos \omega (m){)}^2\\ & & +{\sin }^2\tilde{\omega }\prod _{m=2,4,\dots }^M(\cos \tilde{\omega }-\cos \omega (m){)}^2\},\end{array}\end{array}$$

where $\tilde{\omega }$ is the angular frequency warped by the first-order all pass filter:

$$\tilde{\omega }=\omega +2{\tan }^{-1}\left(\frac{\alpha \sin \omega }{1-\alpha \cos \omega }\right).$$

[1] A. V. Oppenheim and D. H. Johnson, “Discrete representation of signals,” Proc. of the IEEE, vol. 60, no. 6, pp. 681-691, 1972.

[2] N. Sugamura and F. Itakura, “Speech data compression by LSP speech analysis-synthesis technique,” Journal of IEICE, vol. J64-A, no. 8, pp. 599-606, 1981.

Public Functions

MelGeneralizedLineSpectralPairsToSpectrum(int num_order, double alpha, double gamma, int fft_length)

Parameters:

• num_order – [in] Order of line spectral pairs, $M$.

• alpha – [in] Alpha, $\alpha$.

• gamma – [in] Gamma, $\gamma$.

• fft_length – [in] FFT length, $L$.

inline int GetNumOrder() const

Returns:

Order of coefficients.

inline double GetAlpha() const

Returns:

Alpha.

inline double GetGamma() const

Returns:

Gamma.

inline int GetFftLength() const

Returns:

FFT length.

inline bool IsValid() const

Returns:

True if this object is valid.

bool Run(const std::vector<double> &line_spectral_pairs, std::vector<double> *spectrum) const

Parameters:

• line_spectral_pairs – [in] $M$-th order line spectral pairs. The first element is linear gain and the other elements are in normalized frequency $(0,\pi )$.

• spectrum – [out] $(L/2+1)$-length log amplitude spectrum.

Returns:

True on success, false on failure.