Mon Mar 19 11:30:27 2012

Asterisk developer's documentation


func_pitchshift.c

Go to the documentation of this file.
00001 /*
00002  * Asterisk -- An open source telephony toolkit.
00003  *
00004  * Copyright (C) 2010, Digium, Inc.
00005  *
00006  * David Vossel <dvossel@digium.com>
00007  *
00008  * See http://www.asterisk.org for more information about
00009  * the Asterisk project. Please do not directly contact
00010  * any of the maintainers of this project for assistance;
00011  * the project provides a web site, mailing lists and IRC
00012  * channels for your use.
00013  *
00014  * This program is free software, distributed under the terms of
00015  * the GNU General Public License Version 2. See the LICENSE file
00016  * at the top of the source tree.
00017  */
00018 
00019 /*! \file
00020  *
00021  * \brief Pitch Shift Audio Effect
00022  *
00023  * \author David Vossel <dvossel@digium.com>
00024  *
00025  * \ingroup functions
00026  */
00027 
00028 /************************* SMB FUNCTION LICENSE *********************************
00029 *
00030 * SYNOPSIS: Routine for doing pitch shifting while maintaining
00031 * duration using the Short Time Fourier Transform.
00032 *
00033 * DESCRIPTION: The routine takes a pitchShift factor value which is between 0.5
00034 * (one octave down) and 2. (one octave up). A value of exactly 1 does not change
00035 * the pitch. num_samps_to_process tells the routine how many samples in indata[0...
00036 * num_samps_to_process-1] should be pitch shifted and moved to outdata[0 ...
00037 * num_samps_to_process-1]. The two buffers can be identical (ie. it can process the
00038 * data in-place). fft_frame_size defines the FFT frame size used for the
00039 * processing. Typical values are 1024, 2048 and 4096. It may be any value <=
00040 * MAX_FRAME_LENGTH but it MUST be a power of 2. osamp is the STFT
00041 * oversampling factor which also determines the overlap between adjacent STFT
00042 * frames. It should at least be 4 for moderate scaling ratios. A value of 32 is
00043 * recommended for best quality. sampleRate takes the sample rate for the signal
00044 * in unit Hz, ie. 44100 for 44.1 kHz audio. The data passed to the routine in
00045 * indata[] should be in the range [-1.0, 1.0), which is also the output range
00046 * for the data, make sure you scale the data accordingly (for 16bit signed integers
00047 * you would have to divide (and multiply) by 32768).
00048 *
00049 * COPYRIGHT 1999-2009 Stephan M. Bernsee <smb [AT] dspdimension [DOT] com>
00050 *
00051 *                        The Wide Open License (WOL)
00052 *
00053 * Permission to use, copy, modify, distribute and sell this software and its
00054 * documentation for any purpose is hereby granted without fee, provided that
00055 * the above copyright notice and this license appear in all source copies.
00056 * THIS SOFTWARE IS PROVIDED "AS IS" WITHOUT EXPRESS OR IMPLIED WARRANTY OF
00057 * ANY KIND. See http://www.dspguru.com/wol.htm for more information.
00058 *
00059 *****************************************************************************/
00060 
00061 /*** MODULEINFO
00062    <support_level>extended</support_level>
00063  ***/
00064 
00065 #include "asterisk.h"
00066 
00067 ASTERISK_FILE_VERSION(__FILE__, "$Revision: 328209 $")
00068 
00069 #include "asterisk/module.h"
00070 #include "asterisk/channel.h"
00071 #include "asterisk/pbx.h"
00072 #include "asterisk/utils.h"
00073 #include "asterisk/audiohook.h"
00074 #include <math.h>
00075 
00076 /*** DOCUMENTATION
00077    <function name="PITCH_SHIFT" language="en_US">
00078       <synopsis>
00079          Pitch shift both tx and rx audio streams on a channel.
00080       </synopsis>
00081       <syntax>
00082          <parameter name="channel direction" required="true">
00083             <para>Direction can be either <literal>rx</literal>, <literal>tx</literal>, or
00084             <literal>both</literal>.  The direction can either be set to a valid floating
00085             point number between 0.1 and 4.0 or one of the enum values listed below. A value
00086             of 1.0 has no effect.  Greater than 1 raises the pitch. Lower than 1 lowers
00087             the pitch.</para>
00088 
00089             <para>The pitch amount can also be set by the following values</para>
00090             <enumlist>
00091                <enum name = "highest" />
00092                <enum name = "higher" />
00093                <enum name = "high" />
00094                <enum name = "low" />
00095                <enum name = "lower" />
00096                <enum name = "lowest" />
00097             </enumlist>
00098          </parameter>
00099       </syntax>
00100       <description>
00101          <para>Examples:</para>
00102          <para>exten => 1,1,Set(PITCH_SHIFT(tx)=highest); raises pitch an octave </para>
00103          <para>exten => 1,1,Set(PITCH_SHIFT(rx)=higher) ; raises pitch more </para>
00104          <para>exten => 1,1,Set(PITCH_SHIFT(both)=high)   ; raises pitch </para>
00105          <para>exten => 1,1,Set(PITCH_SHIFT(rx)=low)    ; lowers pitch </para>
00106          <para>exten => 1,1,Set(PITCH_SHIFT(tx)=lower)  ; lowers pitch more </para>
00107          <para>exten => 1,1,Set(PITCH_SHIFT(both)=lowest) ; lowers pitch an octave </para>
00108 
00109          <para>exten => 1,1,Set(PITCH_SHIFT(rx)=0.8)    ; lowers pitch </para>
00110          <para>exten => 1,1,Set(PITCH_SHIFT(tx)=1.5)    ; raises pitch </para>
00111       </description>
00112    </function>
00113  ***/
00114 
00115 #ifndef M_PI
00116 #define M_PI 3.14159265358979323846
00117 #endif
00118 #define MAX_FRAME_LENGTH 256
00119 
00120 #define HIGHEST 2
00121 #define HIGHER 1.5
00122 #define HIGH 1.25
00123 #define LOW .85
00124 #define LOWER .7
00125 #define LOWEST .5
00126 
00127 struct fft_data {
00128    float in_fifo[MAX_FRAME_LENGTH];
00129    float out_fifo[MAX_FRAME_LENGTH];
00130    float fft_worksp[2*MAX_FRAME_LENGTH];
00131    float last_phase[MAX_FRAME_LENGTH/2+1];
00132    float sum_phase[MAX_FRAME_LENGTH/2+1];
00133    float output_accum[2*MAX_FRAME_LENGTH];
00134    float ana_freq[MAX_FRAME_LENGTH];
00135    float ana_magn[MAX_FRAME_LENGTH];
00136    float syn_freq[MAX_FRAME_LENGTH];
00137    float sys_magn[MAX_FRAME_LENGTH];
00138    long gRover;
00139    float shift_amount;
00140 };
00141 
00142 struct pitchshift_data {
00143    struct ast_audiohook audiohook;
00144 
00145    struct fft_data rx;
00146    struct fft_data tx;
00147 };
00148 
00149 static void smb_fft(float *fft_buffer, long fft_frame_size, long sign);
00150 static void smb_pitch_shift(float pitchShift, long num_samps_to_process, long fft_frame_size, long osamp, float sample_rate, int16_t *indata, int16_t *outdata, struct fft_data *fft_data);
00151 static int pitch_shift(struct ast_frame *f, float amount, struct fft_data *fft_data);
00152 
00153 static void destroy_callback(void *data)
00154 {
00155    struct pitchshift_data *shift = data;
00156 
00157    ast_audiohook_destroy(&shift->audiohook);
00158    ast_free(shift);
00159 };
00160 
00161 static const struct ast_datastore_info pitchshift_datastore = {
00162    .type = "pitchshift",
00163    .destroy = destroy_callback
00164 };
00165 
00166 static int pitchshift_cb(struct ast_audiohook *audiohook, struct ast_channel *chan, struct ast_frame *f, enum ast_audiohook_direction direction)
00167 {
00168    struct ast_datastore *datastore = NULL;
00169    struct pitchshift_data *shift = NULL;
00170 
00171 
00172    if (!f) {
00173       return 0;
00174    }
00175    if ((audiohook->status == AST_AUDIOHOOK_STATUS_DONE) ||
00176       (f->frametype != AST_FRAME_VOICE) ||
00177       ((f->subclass.codec != AST_FORMAT_SLINEAR) &&
00178       (f->subclass.codec != AST_FORMAT_SLINEAR16))) {
00179       return -1;
00180    }
00181 
00182    if (!(datastore = ast_channel_datastore_find(chan, &pitchshift_datastore, NULL))) {
00183       return -1;
00184    }
00185 
00186    shift = datastore->data;
00187 
00188    if (direction == AST_AUDIOHOOK_DIRECTION_WRITE) {
00189       pitch_shift(f, shift->tx.shift_amount, &shift->tx);
00190    } else {
00191       pitch_shift(f, shift->rx.shift_amount, &shift->rx);
00192    }
00193 
00194    return 0;
00195 }
00196 
00197 static int pitchshift_helper(struct ast_channel *chan, const char *cmd, char *data, const char *value)
00198 {
00199    struct ast_datastore *datastore = NULL;
00200    struct pitchshift_data *shift = NULL;
00201    int new = 0;
00202    float amount = 0;
00203 
00204    ast_channel_lock(chan);
00205    if (!(datastore = ast_channel_datastore_find(chan, &pitchshift_datastore, NULL))) {
00206       ast_channel_unlock(chan);
00207 
00208       if (!(datastore = ast_datastore_alloc(&pitchshift_datastore, NULL))) {
00209          return 0;
00210       }
00211       if (!(shift = ast_calloc(1, sizeof(*shift)))) {
00212          ast_datastore_free(datastore);
00213          return 0;
00214       }
00215 
00216       ast_audiohook_init(&shift->audiohook, AST_AUDIOHOOK_TYPE_MANIPULATE, "pitch_shift");
00217       shift->audiohook.manipulate_callback = pitchshift_cb;
00218       datastore->data = shift;
00219       new = 1;
00220    } else {
00221       ast_channel_unlock(chan);
00222       shift = datastore->data;
00223    }
00224 
00225 
00226    if (!strcasecmp(value, "highest")) {
00227       amount = HIGHEST;
00228    } else if (!strcasecmp(value, "higher")) {
00229       amount = HIGHER;
00230    } else if (!strcasecmp(value, "high")) {
00231       amount = HIGH;
00232    } else if (!strcasecmp(value, "lowest")) {
00233       amount = LOWEST;
00234    } else if (!strcasecmp(value, "lower")) {
00235       amount = LOWER;
00236    } else if (!strcasecmp(value, "low")) {
00237       amount = LOW;
00238    } else {
00239       if (!sscanf(value, "%30f", &amount) || (amount <= 0) || (amount > 4)) {
00240          goto cleanup_error;
00241       }
00242    }
00243 
00244    if (!strcasecmp(data, "rx")) {
00245       shift->rx.shift_amount = amount;
00246    } else if (!strcasecmp(data, "tx")) {
00247       shift->tx.shift_amount = amount;
00248    } else if (!strcasecmp(data, "both")) {
00249       shift->rx.shift_amount = amount;
00250       shift->tx.shift_amount = amount;
00251    } else {
00252       goto cleanup_error;
00253    }
00254 
00255    if (new) {
00256       ast_channel_lock(chan);
00257       ast_channel_datastore_add(chan, datastore);
00258       ast_channel_unlock(chan);
00259       ast_audiohook_attach(chan, &shift->audiohook);
00260    }
00261 
00262    return 0;
00263 
00264 cleanup_error:
00265 
00266    ast_log(LOG_ERROR, "Invalid argument provided to the %s function\n", cmd);
00267    if (new) {
00268       ast_datastore_free(datastore);
00269    }
00270    return -1;
00271 }
00272 
00273 static void smb_fft(float *fft_buffer, long fft_frame_size, long sign)
00274 {
00275    float wr, wi, arg, *p1, *p2, temp;
00276    float tr, ti, ur, ui, *p1r, *p1i, *p2r, *p2i;
00277    long i, bitm, j, le, le2, k;
00278 
00279    for (i = 2; i < 2 * fft_frame_size - 2; i += 2) {
00280       for (bitm = 2, j = 0; bitm < 2 * fft_frame_size; bitm <<= 1) {
00281          if (i & bitm) {
00282             j++;
00283          }
00284          j <<= 1;
00285       }
00286       if (i < j) {
00287          p1 = fft_buffer + i; p2 = fft_buffer + j;
00288          temp = *p1; *(p1++) = *p2;
00289          *(p2++) = temp; temp = *p1;
00290          *p1 = *p2; *p2 = temp;
00291       }
00292    }
00293    for (k = 0, le = 2; k < (long) (log(fft_frame_size) / log(2.) + .5); k++) {
00294       le <<= 1;
00295       le2 = le>>1;
00296       ur = 1.0;
00297       ui = 0.0;
00298       arg = M_PI / (le2>>1);
00299       wr = cos(arg);
00300       wi = sign * sin(arg);
00301       for (j = 0; j < le2; j += 2) {
00302          p1r = fft_buffer+j; p1i = p1r + 1;
00303          p2r = p1r + le2; p2i = p2r + 1;
00304          for (i = j; i < 2 * fft_frame_size; i += le) {
00305             tr = *p2r * ur - *p2i * ui;
00306             ti = *p2r * ui + *p2i * ur;
00307             *p2r = *p1r - tr; *p2i = *p1i - ti;
00308             *p1r += tr; *p1i += ti;
00309             p1r += le; p1i += le;
00310             p2r += le; p2i += le;
00311          }
00312          tr = ur * wr - ui * wi;
00313          ui = ur * wi + ui * wr;
00314          ur = tr;
00315       }
00316    }
00317 }
00318 
00319 static void smb_pitch_shift(float pitchShift, long num_samps_to_process, long fft_frame_size, long osamp, float sample_rate, int16_t *indata, int16_t *outdata, struct fft_data *fft_data)
00320 {
00321    float *in_fifo = fft_data->in_fifo;
00322    float *out_fifo = fft_data->out_fifo;
00323    float *fft_worksp = fft_data->fft_worksp;
00324    float *last_phase = fft_data->last_phase;
00325    float *sum_phase = fft_data->sum_phase;
00326    float *output_accum = fft_data->output_accum;
00327    float *ana_freq = fft_data->ana_freq;
00328    float *ana_magn = fft_data->ana_magn;
00329    float *syn_freq = fft_data->syn_freq;
00330    float *sys_magn = fft_data->sys_magn;
00331 
00332    double magn, phase, tmp, window, real, imag;
00333    double freq_per_bin, expct;
00334    long i,k, qpd, index, in_fifo_latency, step_size, fft_frame_size2;
00335 
00336    /* set up some handy variables */
00337    fft_frame_size2 = fft_frame_size / 2;
00338    step_size = fft_frame_size / osamp;
00339    freq_per_bin = sample_rate / (double) fft_frame_size;
00340    expct = 2. * M_PI * (double) step_size / (double) fft_frame_size;
00341    in_fifo_latency = fft_frame_size-step_size;
00342 
00343    if (fft_data->gRover == 0) {
00344       fft_data->gRover = in_fifo_latency;
00345    }
00346 
00347    /* main processing loop */
00348    for (i = 0; i < num_samps_to_process; i++){
00349 
00350       /* As long as we have not yet collected enough data just read in */
00351       in_fifo[fft_data->gRover] = indata[i];
00352       outdata[i] = out_fifo[fft_data->gRover - in_fifo_latency];
00353       fft_data->gRover++;
00354 
00355       /* now we have enough data for processing */
00356       if (fft_data->gRover >= fft_frame_size) {
00357          fft_data->gRover = in_fifo_latency;
00358 
00359          /* do windowing and re,im interleave */
00360          for (k = 0; k < fft_frame_size;k++) {
00361             window = -.5 * cos(2. * M_PI * (double) k / (double) fft_frame_size) + .5;
00362             fft_worksp[2*k] = in_fifo[k] * window;
00363             fft_worksp[2*k+1] = 0.;
00364          }
00365 
00366          /* ***************** ANALYSIS ******************* */
00367          /* do transform */
00368          smb_fft(fft_worksp, fft_frame_size, -1);
00369 
00370          /* this is the analysis step */
00371          for (k = 0; k <= fft_frame_size2; k++) {
00372 
00373             /* de-interlace FFT buffer */
00374             real = fft_worksp[2*k];
00375             imag = fft_worksp[2*k+1];
00376 
00377             /* compute magnitude and phase */
00378             magn = 2. * sqrt(real * real + imag * imag);
00379             phase = atan2(imag, real);
00380 
00381             /* compute phase difference */
00382             tmp = phase - last_phase[k];
00383             last_phase[k] = phase;
00384 
00385             /* subtract expected phase difference */
00386             tmp -= (double) k * expct;
00387 
00388             /* map delta phase into +/- Pi interval */
00389             qpd = tmp / M_PI;
00390             if (qpd >= 0) {
00391                qpd += qpd & 1;
00392             } else {
00393                qpd -= qpd & 1;
00394             }
00395             tmp -= M_PI * (double) qpd;
00396 
00397             /* get deviation from bin frequency from the +/- Pi interval */
00398             tmp = osamp * tmp / (2. * M_PI);
00399 
00400             /* compute the k-th partials' true frequency */
00401             tmp = (double) k * freq_per_bin + tmp * freq_per_bin;
00402 
00403             /* store magnitude and true frequency in analysis arrays */
00404             ana_magn[k] = magn;
00405             ana_freq[k] = tmp;
00406 
00407          }
00408 
00409          /* ***************** PROCESSING ******************* */
00410          /* this does the actual pitch shifting */
00411          memset(sys_magn, 0, fft_frame_size * sizeof(float));
00412          memset(syn_freq, 0, fft_frame_size * sizeof(float));
00413          for (k = 0; k <= fft_frame_size2; k++) {
00414             index = k * pitchShift;
00415             if (index <= fft_frame_size2) {
00416                sys_magn[index] += ana_magn[k];
00417                syn_freq[index] = ana_freq[k] * pitchShift;
00418             }
00419          }
00420 
00421          /* ***************** SYNTHESIS ******************* */
00422          /* this is the synthesis step */
00423          for (k = 0; k <= fft_frame_size2; k++) {
00424 
00425             /* get magnitude and true frequency from synthesis arrays */
00426             magn = sys_magn[k];
00427             tmp = syn_freq[k];
00428 
00429             /* subtract bin mid frequency */
00430             tmp -= (double) k * freq_per_bin;
00431 
00432             /* get bin deviation from freq deviation */
00433             tmp /= freq_per_bin;
00434 
00435             /* take osamp into account */
00436             tmp = 2. * M_PI * tmp / osamp;
00437 
00438             /* add the overlap phase advance back in */
00439             tmp += (double) k * expct;
00440 
00441             /* accumulate delta phase to get bin phase */
00442             sum_phase[k] += tmp;
00443             phase = sum_phase[k];
00444 
00445             /* get real and imag part and re-interleave */
00446             fft_worksp[2*k] = magn * cos(phase);
00447             fft_worksp[2*k+1] = magn * sin(phase);
00448          }
00449 
00450          /* zero negative frequencies */
00451          for (k = fft_frame_size + 2; k < 2 * fft_frame_size; k++) {
00452             fft_worksp[k] = 0.;
00453          }
00454 
00455          /* do inverse transform */
00456          smb_fft(fft_worksp, fft_frame_size, 1);
00457 
00458          /* do windowing and add to output accumulator */
00459          for (k = 0; k < fft_frame_size; k++) {
00460             window = -.5 * cos(2. * M_PI * (double) k / (double) fft_frame_size) + .5;
00461             output_accum[k] += 2. * window * fft_worksp[2*k] / (fft_frame_size2 * osamp);
00462          }
00463          for (k = 0; k < step_size; k++) {
00464             out_fifo[k] = output_accum[k];
00465          }
00466 
00467          /* shift accumulator */
00468          memmove(output_accum, output_accum+step_size, fft_frame_size * sizeof(float));
00469 
00470          /* move input FIFO */
00471          for (k = 0; k < in_fifo_latency; k++) {
00472             in_fifo[k] = in_fifo[k+step_size];
00473          }
00474       }
00475    }
00476 }
00477 
00478 static int pitch_shift(struct ast_frame *f, float amount, struct fft_data *fft)
00479 {
00480    int16_t *fun = (int16_t *) f->data.ptr;
00481    int samples;
00482 
00483    /* an amount of 1 has no effect */
00484    if (!amount || amount == 1 || !fun || (f->samples % 32)) {
00485       return 0;
00486    }
00487    for (samples = 0; samples < f->samples; samples += 32) {
00488       smb_pitch_shift(amount, 32, MAX_FRAME_LENGTH, 32, ast_format_rate(f->subclass.codec), fun+samples, fun+samples, fft);
00489    }
00490 
00491    return 0;
00492 }
00493 
00494 static struct ast_custom_function pitch_shift_function = {
00495    .name = "PITCH_SHIFT",
00496    .write = pitchshift_helper,
00497 };
00498 
00499 static int unload_module(void)
00500 {
00501    return ast_custom_function_unregister(&pitch_shift_function);
00502 }
00503 
00504 static int load_module(void)
00505 {
00506    int res = ast_custom_function_register(&pitch_shift_function);
00507    return res ? AST_MODULE_LOAD_DECLINE : AST_MODULE_LOAD_SUCCESS;
00508 }
00509 
00510 AST_MODULE_INFO_STANDARD(ASTERISK_GPL_KEY, "Audio Effects Dialplan Functions");

Generated on Mon Mar 19 11:30:27 2012 for Asterisk - The Open Source Telephony Project by  doxygen 1.4.7