hssf.h File Reference

#include <stdio.h>
#include "hs.h"

Include dependency graph for hssf.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures
struct	TSF

Functions
int	Sign (double)
void	S_F_b (TSF &SF, atom **Partials)
double	P2_DelFtr (double **d, int L, int K, double **x, double F)
double	P3_DelSrc (double **d, int L, int M, int K, double **x, double **f, double F)
int	SF_SV (double **h, int L, int M, int K, double **a, double **f, double F, double theta, double ep, int maxiter)
double	S_F_SV (int M, atom **Partials, double *logA0C, double *lp, int P, int &K, double **h, double *avgh, double **b, double *avgb, double F=0.005, int FScaleMode=0, double theta=0.5, double Fs=44100)
void	SF_SV_cf (double *h, double **b, int L, int M, int K, double **a, double **f, double F, double ep, int maxiter)
double	S_F_SV_cf (int afres, double *LogAF, double *LogAS, int Fr, int M, atom **Partials, double *A0C, double *lp, int P, int &K, double **h, double **b, double F=0.005, int FScaleMode=0, double Fs=44100)
int	SF_FB (double *hl, int M, int K, double **al, double **fl, double F, int LMode)
double	S_F_FB (int M, atom **Partials, double *logA0C, double *lp, int P, int &K, double **h, double *avgh, double **b, double *avgb, double F=0.005, int FScaleMode=0, double Fs=44100)
void	AnalyzeSF_1 (THS &HS, TSF &SF, double sps, double offst)
void	AnalyzeSF_2 (THS &HS, TSF &SF, double *&cyclefrs, double *&cyclefs, double sps, int *cyclefrcount=0, int SFMode=0, double SFF=0.01, int SFFScale=0, double SFtheta=0)
void	AnalyzeSF (THS &HS, TSF &SF, double *&cyclefrs, double *&cyclefs, double sps, double offst, int *cyclefrcount=0, int SFMode=0, double SFF=0.01, int SFFScale=0, double SFtheta=0)
void	SynthesizeSF (THS *HS, TSF *SF, double sps)

Variables
const double	Amel =1127.0104803341574386544633680278
const bool	useA0 =true

Detailed Description

• source-filter modeling for harmonic sinusoids

Further reading: Wen X. and M. Sandler, "Source-filter modeling in sinusoid domain," in Proc. AES 126th Convention, Munich, 2009.

Function Documentation

void AnalyzeSF	(	THS &	HS,
		TSF &	SF,
		double *&	cyclefrs,
		double *&	cyclefs,
		double	sps,
		double	offst,
		int *	cyclefrcount,
		int	SFMode,
		double	SFF,
		int	SFFScale,
		double	SFtheta
	)

function AnalyzeSF: wrapper function.

In: HS: a harmonic sinusoid sps: sampling rate ("samples per second") offst: hop size, the interval between adjacent harmonic atoms of HS SFMode: specifies source-filter estimation criterion, 0=FB (filter bank), 1=SV (slow variation) SFF: filter response sampling interval SFScale: set if filter sampling uses mel scale SFtheta: balancing factor of amplitude and frequency variations, needed for SV approach Out: SF: a TSF object estimated from HS. cyclefrcount: number of cycles cyclefrs[cyclefrcount], cyclefs[cyclefrcount]: average time (in frames) and frequency of each cycle

No return value.

void AnalyzeSF_1	(	THS &	HS,
		TSF &	SF,
		double	sps,
		double	offst
	)

function AnalyzeSF_1: first stage of source-filter model estimation, in which the duration of a HS is segmented into what is equivalent to "cycles" in vibrato analysis.

In: HS: a harmonic sinusoid sps: sampling rate ("samples per second") offst: hop size, the interval between adjacent harmonic atoms of HS Out: SF: a TSF object partially updated, particularly lp[P].

No return value.

void AnalyzeSF_2	(	THS &	HS,
		TSF &	SF,
		double *&	cyclefrs,
		double *&	cyclefs,
		double	sps,
		int *	cyclefrcount,
		int	SFMode,
		double	SFF,
		int	SFFScale,
		double	SFtheta
	)

function AnalyzeSF_2: second stage of source-filter model estimation, in which the HS is demodulated and its source-filter model is estimated.

In: HS: a harmonic sinusoid SF: a TSF object with valid segmentation data (lp[P]) of HS sps: sampling rate ("samples per second") SFMode: specifies source-filter estimation criterion, 0=FB (filter bank), 1=SV (slow variation) SFF: filter response sampling interval SFScale: set if filter sampling uses mel scale SFtheta: balancing factor of amplitude and frequency variations, needed for SV approach Out: SF: the TSF object fully estimated. cyclefrcount: number of cycles cyclefrs[cyclefrcount], cyclefs[cyclefrcount]: average time (in frames) and frequency of each cycle

No return value.

double P2_DelFtr	(	double **	d,
		int	L,
		int	K,
		double **	x,
		double	F
	)

function P2_DelFtr: internal procedure P2 of slow-variation SF estimator. See "further reading", pp.8.

In: x[L][K+2]: filter coefficients F: filter response sampling interval Out: d[L][K+2]: derivative of total filter variation against filter coefficients.

Returns the inner product of x[][] and d[][].

double P3_DelSrc	(	double **	d,
		int	L,
		int	M,
		int	K,
		double **	x,
		double **	f,
		double	F
	)

function P3_DelSec: internal procedure P3 of slow-variation SF estimator. See "further reading", pp.9.

In: x[L][K+2]: filter coefficients f[L][M]: partial frequencies F: filter response sampling interval Out: d[L][K+2]: derivative of total source variation against filter coefficients.

Returns the inner product of x[][] and d[][].

void S_F_b	(	TSF &	SF,
		atom **	Partials
	)

function S_F_b: wrapper function

In: Partials: HS partials SF: TSF object containing valid filter coefficients Out: SF: TSF object with source coefficients updated

No return value.

double S_F_FB	(	int	M,
		atom **	Partials,
		double *	logA0C,
		double *	lp,
		int	P,
		int &	K,
		double **	h,
		double *	avgh,
		double **	b,
		double *	avgb,
		double	F,
		int	FScaleMode,
		double	Fs
	)

function S_F_b: filter-bank method for estimating source-filter model. This is a wrapper function of SF_FB() that transfers and scales values, etc.

In: Partials[M][Fr]: HS partials. Fr is not specified but is assumed no less then lp[P-1]. lp[P]: temporal segmentation points, in frames. logA0C[Fr]: amplitude carrier F: filter response sampling interval FScaleMode: set if mel scale is used as frequency scale Fs: sampling frequency, effective only when FScaleMode is set. Out: K h[L][K+2], aveh[K+2]: filter coefficients for L frames from ceil(lp[0]) to ceil(lp[P-1]), and their averages b[L][M], avgb[M]: source coefficients for L frames from ceil(lp[0]) to ceil(lp[P-1]), and their averages

Returns 0.

double S_F_SV	(	int	M,
		atom **	Partials,
		double *	logA0C,
		double *	lp,
		int	P,
		int &	K,
		double **	h,
		double *	avgh,
		double **	b,
		double *	avgb,
		double	F,
		int	FScaleMode,
		double	theta,
		double	Fs
	)

function S_F_SV: slow-variation method for estimating source-filter model. This is a wrapper function of SF_SV() that transfers and scales values, etc.

In: Partials[M][Fr]: HS partials. Fr is not specified but is assumed no less then lp[P-1]. lp[P]: temporal segmentation points, in frames. logA0C[Fr]: amplitude carrier theta: balancing factor of amplitude and frequency variations F: filter response sampling interval FScaleMode: set if mel scale is used as frequency scale Fs: sampling frequency, effective only when FScaleMode is set. Out: K h[L][K+2], aveh[K+2]: filter coefficients for L frames from ceil(lp[0]) to ceil(lp[P-1]), and their averages b[L][M], avgb[M]: source coefficients for L frames from ceil(lp[0]) to ceil(lp[P-1]), and their averages Returns 0.

double S_F_SV_cf	(	int	afres,
		double *	LogAF,
		double *	LogAS,
		int	Fr,
		int	M,
		atom **	Partials,
		double *	A0C,
		double *	lp,
		int	P,
		int &	K,
		double **	h,
		double **	b,
		double	F,
		int	FScaleMode,
		double	Fs
	)

function S_F_SV_cf: slow-variation method for estimating source-(constant)filter model. This is a wrapper function of SF_SV_cf() that transfers and scales values, as well as converting results to source filter format used by the vibrato class TVo.

In: Partials[M][Fr]: HS partials. Fr is not specified but is assumed no less then lp[P-1]. lp[P]: temporal segmentation points, in frames. A0C[Fr]: amplitude carrier (linear, not dB) F: filter response sampling interval FScaleMode: set if mel scale is used as frequency scale Fs: sampling frequency, effective only when FScaleMode is set. Out: K h[L][K+2], aveh[K+2]: filter coefficients for L frames from ceil(lp[0]) to ceil(lp[P-1]), and their averages b[L][M], avgb[M]: source coefficients for L frames from ceil(lp[0]) to ceil(lp[P-1]), and their averages LogAF[afres/2]: TVo-format filter response LogAS[M]: TVo-format source response

Returns 0.

int SF_FB	(	double *	hl,
		int	M,
		int	K,
		double **	al,
		double **	fl,
		double	F,
		int	LMode
	)

function SF_FB: computes filter coefficients with filter-bank method: single frame. See "further reading", pp.12, P5.

In: al[][M], fl[][M]: partial amplitudes and frequencies, al[0] and fl[0] are of current frame F: filter response sampling interval LMode: 1: current frame is the first frame; -1: the last frame; 0: neither first nor last frame Out: hl[K+2]: filter coefficients integrated at current frame

Returns K.

int SF_SV	(	double **	h,
		int	L,
		int	M,
		int	K,
		double **	a,
		double **	f,
		double	F,
		double	theta,
		double	ep,
		int	maxiter
	)

function SF_SV: CG method for estimation source-filter model using slow-variation criterion. See "further reading", pp.9, boxed algorithm P4.

In: a[L][M], f[L][M]: partial amplitudes and frequencies F: filter response sampling interval theta: balancing factor of amplitude and frequency variations ep: convergence test threshold maxiter: maximal number of iterates Out: h[L][K+2]: filter coefficients

Returns K.

void SF_SV_cf	(	double *	h,
		double **	b,
		int	L,
		int	M,
		int	K,
		double **	a,
		double **	f,
		double	F,
		double	ep,
		int	maxiter
	)

function SF_SV_cf: CG method for estimation source-(constant)filter model by slow-variation criterion.

In: a[L][M], f[L][M]: partial amplitudes and frequencies F: filter response sampling interval ep: convergence test threshold maxiter: maximal number of iterates Out: h[K+2]: filter coefficients b[L][M]: source coefficients

No reutrn value.

int Sign

(

double

x

)

function Sign: sign function

In: x: a floating-point number

Returns 0 if x=0, 1 if x>0, -1 if x<0.

void SynthesizeSF	(	THS *	HS,
		TSF *	SF,
		double	sps
	)

function SynthesizeSF: synthesizes a harmonic sinusoid representation from a TSF object.

In: SF: a TSF object sps: sampling rate ("samples per second") Out: HS: teh synthesized HS.

No return value.