⭐ 欢迎来到虫虫下载站! | 📦 资源下载 📁 资源专辑 ℹ️ 关于我们
⭐ 虫虫下载站

📄 tonal.c

📁 ISO mp3 sources (distribution 10) Layer 1/2/3, C Source, 512 k Sources of the Mpeg 1,2 layer 1,2
💻 C
📖 第 1 页 / 共 4 页
字号:
/********************************************************************** * ISO MPEG Audio Subgroup Software Simulation Group (1996) * ISO 13818-3 MPEG-2 Audio Multichannel Encoder * * $Id: tonal.c 1.8 1996/02/12 07:13:35 rowlands Exp $ * * $Log: tonal.c $ * Revision 1.8  1996/02/12 07:13:35  rowlands * Release following Munich meeting * * Revision 1.5.2.1  1995/11/06  04:19:12  rowlands * Received from Uwe Felderhoff (IRT) * * Revision 1.7  1995/08/14  08:06:37  tenkate * ML-LSF added Warner ten Kate 7/8/95 (Philips) * II_psycho_one() split into II_psycho_one() for audio data and * II_psycho_one_ml() for MultiLingual data. * Variables crit_band, cbound and sub_size has been made local to * these functions. * Tables "cb" and "th" are copied from LSF-directory. * ltg has found its ml counterpart. * * Revision 1.6  1995/07/31  07:47:50  tenkate * bugs correction (if center==1 in Psycho_One), 25/07/95 WtK * * Revision 1.4.2.1  1995/06/16  03:46:42  rowlands * Input from Susanne Ritscher (IRT) * **********************************************************************//********************************************************************** *   date   programmers         comment                               * * 2/25/91  Douglas Wong        start of version 1.1 records          * * 3/06/91  Douglas Wong        rename: setup.h to endef.h            * *                              updated I_psycho_one and II_psycho_one* * 3/11/91  W. J. Carter        Added Douglas Wong's updates dated    * *                              3/9/91 for I_Psycho_One() and for     * *                              II_Psycho_One().                      * * 5/10/91  W. Joseph Carter    Ported to Macintosh and Unix.         * *                              Located and fixed numerous software   * *                              bugs and table data errors.           * * 6/11/91  Davis Pan           corrected several bugs                * *                              based on comments from H. Fuchs       * * 01jul91  dpwe (Aware Inc.)   Made pow() args float                 * *                              Removed logical bug in I_tonal_label: * *                              Sometimes *tone returned == STOP      * * 7/10/91  Earle Jennings      no change necessary in port to MsDos  * * 11sep91  dpwe@aware.com      Subtracted 90.3dB from II_f_f_t peaks * * 10/1/91  Peter W. Farrett    Updated II_Psycho_One(),I_Psycho_One()* *				to include comments.		      * *11/29/91  Masahiro Iwadare    Bug fix regarding POWERNORM           * *                              fixed several other miscellaneous bugs* * 2/11/92  W. Joseph Carter    Ported new code to Macintosh.  Most   * *                              important fixes involved changing     * *                              16-bit ints to long or unsigned in    * *                              bit alloc routines for quant of 65535 * *                              and passing proper function args.     * *                              Removed "Other Joint Stereo" option   * *                              and made bitrate be total channel     * *                              bitrate, irrespective of the mode.    * *                              Fixed many small bugs & reorganized.  * * 2/12/92  Masahiro Iwadare    Fixed some potential bugs in          * *          Davis Pan           subsampling()                         * * 2/25/92  Masahiro Iwadare    Fixed some more potential bugs        * * 92-11-06 Soren H. Nielsen	Corrected power calculation in I_ and * *				II_f_f_t.		              *	 ********************************************************************** *                                                                    * *                                                                    * *  MPEG/audio Phase 2 coding/decoding multichannel                   * *                                                                    * *  7/27/93        Susanne Ritscher,  IRT Munich                      * *  8/27/93        Susanne Ritscher, IRT Munich                       * *                 Channel-Switching is working                       * *  9/1/93         Susanne Ritscher,  IRT Munich                      * *                 all channels normalized                            * *                                                                    * *  9/20/93        channel-switching is only performed at a           * *                 certain limit of TC_ALLOC dB, which is included    * *                 in encoder.h                                       * *                                                                    * * 10/18/93        seperated smr and ltmin                            * *                                                                    * *  Version 1.0                                                       * *                                                                    * *  07/12/94       Susanne Ritscher,  IRT Munich                      * *                 Tel: +49 89 32399 458                              * *                 Fax: +49 89 32399 415                              * *                                                                    * *  Version 1.1                                                       * *                                                                    * *  02/23/95	   Susanne Ritscher,  IRT Munich                      * *                 corrected some bugs                                * *                 extension bitstream is working                     * *                                                                    * **********************************************************************/#define TONAL_WIDTH 0.1 /* When more than one tonal component is within                           this width in Bark, the weaker one(s) are                           eliminated */#include "common.h"#include "encoder.h"/**********************************************************************//*/*        This module implements the psychoacoustic model I for the/* MPEG encoder layer II. It uses simplified tonal and noise masking/* threshold analysis to generate SMR for the encoder bit allocation/* routine./*/**********************************************************************/int read_crit_band(int lay, int freq) { int crit_band; FILE *fp; char r[16], t[80]; strcpy(r, "2cb1"); r[0] = (char) lay + '0'; r[3] = (char) freq + '0'; if( !(fp = OpenTableFile(r)) ){       /* check boundary values */    printf("Please check %s boundary table\n",r);    exit(0); } fgets(t,80,fp); sscanf(t,"%d\n",&crit_band); fclose(fp); return(crit_band);}        void read_cbound(int lay, int freq, int crit_band, int *cbound)  /* this function reads in critical band boundaries */{ int i,j,k; FILE *fp; char r[16], t[80]; strcpy(r, "2cb1"); r[0] = (char) lay + '0'; r[3] = (char) freq + '0'; if( !(fp = OpenTableFile(r)) ){       /* check boundary values */    printf("Please check %s boundary table\n",r);    exit(0); } fgets(t,80,fp);               /* skip input for critical bands */ sscanf(t,"%d\n",&k); for(i=0;i<crit_band;i++){     fgets(t,80,fp);    sscanf(t,"%d %d\n",&j, &k);    if(i==j) cbound[j] = k;    else {                     /* error */       printf("Please check index %d in cbound table %s\n",i,r);       exit(0);    } } fclose(fp);}        void read_freq_band(int *sub_size, g_ptr *ltg, int lay, int freq)   /* this function reads in frequency bands and bark values  */{ int i,j, k; double a,b,c; FILE *fp; char r[16], t[80]; strcpy(r, "2th1"); r[0] = (char) lay + '0'; r[3] = (char) freq + '0'; if( !(fp = OpenTableFile(r)) ){   /* check freq. values  */    printf("Please check frequency and cband table %s\n",r);    exit(0); } fgets(t,80,fp);              /* read input for freq. subbands */ sscanf(t,"%d\n",sub_size); *ltg = (g_ptr) mem_alloc (sizeof(g_thres) * (*sub_size), "ltg"); (*ltg)[0].line = 0;          /* initialize global masking threshold */ (*ltg)[0].bark = 0; (*ltg)[0].hear = 0; for(i=1;i<(*sub_size);i++){    /* continue to read freq. subband */    fgets(t,80,fp);          /* and assign                     */    sscanf(t,"%d %d %lf %lf\n",&j, &k, &b, &c);    if(i == j){       (*ltg)[j].line = k;       (*ltg)[j].bark = b;       (*ltg)[j].hear = c;    }    else {                   /* error */       printf("Please check index %d in freq-cb table %s\n",i,r);       exit(0);    } } fclose(fp);}void make_map (int sub_size, mask *power, g_thres *ltg)/* this function calculates the global masking threshold */{    int i, j;       for (i = 1; i < sub_size; i++)	for (j = ltg[i-1].line; j <= ltg[i].line; j++)	    power[j].map = i;}double add_db(double a, double b){ a = pow(10.0,a/10.0); b = pow(10.0,b/10.0); return 10 * log10(a+b);}/****************************************************************//*/*        Fast Fourier transform of the input samples./*/****************************************************************/void II_f_f_t (double *sample, mask *power) /* this function calculates an */					    /* FFT analysis for the freq.  */					    /* domain                      */{    int i,j,k,ll,l=0;    int ip, le, le1;    double t_r, t_i, u_r, u_i;    static int M, MM1, init = 0, N, NV2, NM1;    double *x_r, *x_i, *energy;    static int *rev;    static double *w_r, *w_i;    x_r = (double *) mem_alloc (sizeof (DFFT), "x_r");    x_i = (double *) mem_alloc (sizeof (DFFT), "x_i");    energy = (double *) mem_alloc (sizeof (DFFT), "energy");    if (!init)    {	rev = (int *) mem_alloc (sizeof (IFFT), "rev");	w_r = (double *) mem_alloc (sizeof (D10), "w_r");	w_i = (double *) mem_alloc (sizeof (D10), "w_i");	M = 10;	MM1 = 9;	N = FFT_SIZE;	NV2 = FFT_SIZE >> 1;	NM1 = FFT_SIZE - 1;	for (ll = 0; ll < M; ll++)	{	    le = 1 << (M-ll);	    le1 = le >> 1;	    w_r[ll] = cos (PI / le1);	    w_i[ll] = -sin (PI / le1);	}	for (i = 0; i < FFT_SIZE; rev[i] = l, i++)	    for (j = 0, l = 0; j < 10; j++)	    {		k = (i >> j) & 1;		l |= (k << 9 - j);                	    }	init = 1;    }    for (i = 0; i < FFT_SIZE; i++)    {       x_r[i] = sample[i];       x_i[i] = energy[i] = 0;    }/*    memcpy ((char *) x_r, (char *) sample, sizeof (double) * FFT_SIZE);*/    for (ll = 0; ll < MM1; ll++)    {	le = 1 << (M - ll);	le1 = le >> 1;	u_r = 1;	u_i = 0;	for (j = 0; j < le1; j++)	{	    for (i = j; i < N; i += le)	    {		ip = i + le1;		t_r = x_r[i] + x_r[ip];		t_i = x_i[i] + x_i[ip];		x_r[ip] = x_r[i] - x_r[ip];		x_i[ip] = x_i[i] - x_i[ip];		x_r[i] = t_r;		x_i[i] = t_i;		t_r = x_r[ip];		x_r[ip] = x_r[ip] * u_r - x_i[ip] * u_i;		x_i[ip] = x_i[ip] * u_r + t_r * u_i;	    }	    t_r = u_r;	    u_r = u_r * w_r[ll] - u_i * w_i[ll];	    u_i = u_i * w_r[ll] + t_r * w_i[ll];	}    }    for (i = 0; i < N; i += 2)    {	ip = i + 1;	t_r = x_r[i] + x_r[ip];	t_i = x_i[i] + x_i[ip];	x_r[ip] = x_r[i] - x_r[ip];	x_i[ip] = x_i[i] - x_i[ip];	x_r[i] = t_r;	x_i[i] = t_i;	energy[i] = x_r[i] * x_r[i] + x_i[i] * x_i[i];    }    for (i = 0; i < FFT_SIZE; i++)	if (i < rev[i])	{	    t_r = energy[i];	    energy[i] = energy[rev[i]];	    energy[rev[i]] = t_r;	}    for (i = 0; i < HAN_SIZE; i++)    {	/* calculate power density spectrum */	if (energy[i] < 1E-20)	    energy[i] = 1E-20;	/* power calculation corrected with a factor 4, both positive	   and negative frequencies exist, 1992-11-06 shn */	power[i].x = 10 * log10 (energy[i] * 4.0) + POWERNORM;	power[i].next = STOP;	power[i].type = FALSE;    }    mem_free ((void **) &x_r);    mem_free ((void **) &x_i);    mem_free ((void **) &energy);}/****************************************************************//*/*         Window the incoming audio signal./*/****************************************************************/void II_hann_win (double *sample)   /* this function calculates a  */				    /* Hann window for PCM (input) */{				    /* samples for a 1024-pt. FFT  */    register int i;    register double sqrt_8_over_3;    static int init = 0;    static double *window;    

⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -