API_LowLvl::lib::preproc::mod

Collaboration diagram for API_LowLvl::lib::preproc::mod:

Data Structures
struct	SprAGCInfo
struct	SprSDFX
struct	SprSDKD

Enumerations
enum	{ SPR_FB_WARP_LIN1, SPR_FB_WARP_LIN2, SPR_FB_NWARPMETHOD }

Functions
int	spr_agc_connect (SprSspDesc *ssp_des, const char *agc_name, void(*func)(SprAGCInfo *), void *ext_info, int query_sz[5])
void	spr_aut2lpc_levinson (float spr_autocorr[], float refl_coeff[], float LPC_filter[], int order, float LPC_filter_old[])
void	spr_autocorr (const float in[], int nparam, float out[], int ncorr)
void	spr_crosscorr (float in[], float in2[], int nparam, float cr[], float cl[], int ncorr)

Detailed Description

Enumeration Type Documentation

anonymous enum

Enumerator
SPR_FB_WARP_LIN1
SPR_FB_WARP_LIN2
SPR_FB_NWARPMETHOD

Function Documentation

int spr_agc_connect	(	SprSspDesc *	ssp_des,
		const char *	agc_name,
		void(*)(SprAGCInfo *)	func,
		void *	ext_info,
		int	query_sz[5]
	)

Install a function func to interprete the histogram for the AGC with name agc_name. This function may update the record volume (or whatever parameter controls the range of the input data), but this must be reported back by setting the gain_adapt in the AGCinfo structure ext_info to 1.0/gain, with gain the linear gain change corresponding to the adjustment of the record volume.
If the query_sz parameter is not NULL, the following information is returned as well: query_sz = {Nsamples_per_frame, nfr_per_fragment, nfrag_per_hist, nskipa, nskips}

Returns

(-1) on failure (no AGC with the given name found, or this AGC had already an update function installed). Otherwise, a 0 is returned.

void spr_aut2lpc_levinson	(	float	spr_autocorr[],
		float	refl_coeff[],
		float	LPC_filter[],
		int	order,
		float	LPC_filter_old[]
	)

Convert autocorrelation coefficients autocorr[0..order] to reflection coefficients refl_coeff[1..order] and lpc filter coefficients LPC_filter1..order. The coefficient zero of all those vectors are filled in as follows:

• LPC_filter[0] = LPC_gain = sqrt(alpha),
• refl_coeff[0] = autocorr[0] (the average energy).

//initialisation:
  LPC_filter[0] = 1;
  alpha(0) = autocorr[0];                       // error signal energy is energy in signal itself iteration for iorder = 1..order
  (each time used, iparam runs from 1..iorder-1)
  LPC_filter_old[iparam] = LPC_filter[iparam]   // save previous order
  sum = Sum(LPC_filter_old[iparam]*autocorr[iorder-iparam])
  refl_coeff[iorder] = -1.0/alpha(iorder-1)*(autocorr[iorder] + sum
  LPC_filter[iorder] = refl_coeff[iorder];
  LPC_filter[iparam] = LPC_filter_old[iparam] + refl_coeff[iorder]LPC_filter_old[iorder-iparam]
  alpha(iorder) = (1-refl_coeff[iorder]^2) alpha(iorder-1)
//termination:
  LPC_filter[0] = sqrt(alpha)                   // LPC gain
  refl_coeff[0] = autocorr[0];

void spr_autocorr	(	const float	in[],
		int	nparam,
		float	out[],
		int	ncorr
	)

Calculate the normalised autocorrelation coeffients from a time domain signal. It computes ncorr+1 autocorrelation coefficients out[0]...out[Lr] on time series in[0]...in[n-1]. Note by JD: gamma removed, adjustment for taking into account the energy scaling when placing a window, is now done in the windowing routines themselves.

out[icorr] = Sum(in[iparam]*in[iparam+icorr])

void spr_crosscorr	(	float	in[],
		float	in2[],
		int	nparam,
		float	cr[],
		float	cl[],
		int	ncorr
	)

Compute ncorr+1 left and right crosscorrelation coefficients cl[0]...[ncorr] and cr[0]...[ncorr] on time series in[0]...in[nparam-1] (move in) and in2[0]...in2[nparam-1].

cr[0] = cl[0];
cr[iparam] = Sum(in[icorr]*in2[icorr+iparam]);
cl[iparam] = Sum(in[icorr]*in2[icorr-iparam])
           = Sum(in[icorr+iparam]*in2[icorr])).