Wed Apr 6 11:29:45 2011

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 #include "asterisk.h"
00062 
00063 ASTERISK_FILE_VERSION(__FILE__, "$Revision: 251262 $")
00064 
00065 #include "asterisk/module.h"
00066 #include "asterisk/channel.h"
00067 #include "asterisk/pbx.h"
00068 #include "asterisk/utils.h"
00069 #include "asterisk/audiohook.h"
00070 #include <math.h>
00071 
00072 /*** DOCUMENTATION
00073    <function name="PITCH_SHIFT" language="en_US">
00074       <synopsis>
00075          Pitch shift both tx and rx audio streams on a channel.
00076       </synopsis>
00077       <syntax>
00078          <parameter name="channel direction" required="true">
00079             <para>Direction can be either <literal>rx</literal>, <literal>tx</literal>, or
00080             <literal>both</literal>.  The direction can either be set to a valid floating
00081             point number between 0.1 and 4.0 or one of the enum values listed below. A value
00082             of 1.0 has no effect.  Greater than 1 raises the pitch. Lower than 1 lowers
00083             the pitch.</para>
00084 
00085             <para>The pitch amount can also be set by the following values</para>
00086             <enumlist>
00087                <enum name = "highest" />
00088                <enum name = "higher" />
00089                <enum name = "high" />
00090                <enum name = "low" />
00091                <enum name = "lower" />
00092                <enum name = "lowest" />
00093             </enumlist>
00094          </parameter>
00095       </syntax>
00096       <description>
00097          <para>Examples:</para>
00098          <para>exten => 1,1,Set(PITCH_SHIFT(tx)=highest); raises pitch an octave </para>
00099          <para>exten => 1,1,Set(PITCH_SHIFT(rx)=higher) ; raises pitch more </para>
00100          <para>exten => 1,1,Set(PITCH_SHIFT(both)=high)   ; raises pitch </para>
00101          <para>exten => 1,1,Set(PITCH_SHIFT(rx)=low)    ; lowers pitch </para>
00102          <para>exten => 1,1,Set(PITCH_SHIFT(tx)=lower)  ; lowers pitch more </para>
00103          <para>exten => 1,1,Set(PITCH_SHIFT(both)=lowest) ; lowers pitch an octave </para>
00104 
00105          <para>exten => 1,1,Set(PITCH_SHIFT(rx)=0.8)    ; lowers pitch </para>
00106          <para>exten => 1,1,Set(PITCH_SHIFT(tx)=1.5)    ; raises pitch </para>
00107       </description>
00108    </function>
00109  ***/
00110 
00111 #ifndef M_PI
00112 #define M_PI 3.14159265358979323846
00113 #endif
00114 #define MAX_FRAME_LENGTH 256
00115 
00116 #define HIGHEST 2
00117 #define HIGHER 1.5
00118 #define HIGH 1.25
00119 #define LOW .85
00120 #define LOWER .7
00121 #define LOWEST .5
00122 
00123 struct fft_data {
00124    float in_fifo[MAX_FRAME_LENGTH];
00125    float out_fifo[MAX_FRAME_LENGTH];
00126    float fft_worksp[2*MAX_FRAME_LENGTH];
00127    float last_phase[MAX_FRAME_LENGTH/2+1];
00128    float sum_phase[MAX_FRAME_LENGTH/2+1];
00129    float output_accum[2*MAX_FRAME_LENGTH];
00130    float ana_freq[MAX_FRAME_LENGTH];
00131    float ana_magn[MAX_FRAME_LENGTH];
00132    float syn_freq[MAX_FRAME_LENGTH];
00133    float sys_magn[MAX_FRAME_LENGTH];
00134    long gRover;
00135    float shift_amount;
00136 };
00137 
00138 struct pitchshift_data {
00139    struct ast_audiohook audiohook;
00140 
00141    struct fft_data rx;
00142    struct fft_data tx;
00143 };
00144 
00145 static void smb_fft(float *fft_buffer, long fft_frame_size, long sign);
00146 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);
00147 static int pitch_shift(struct ast_frame *f, float amount, struct fft_data *fft_data);
00148 
00149 static void destroy_callback(void *data)
00150 {
00151    struct pitchshift_data *shift = data;
00152 
00153    ast_audiohook_destroy(&shift->audiohook);
00154    ast_free(shift);
00155 };
00156 
00157 static const struct ast_datastore_info pitchshift_datastore = {
00158    .type = "pitchshift",
00159    .destroy = destroy_callback
00160 };
00161 
00162 static int pitchshift_cb(struct ast_audiohook *audiohook, struct ast_channel *chan, struct ast_frame *f, enum ast_audiohook_direction direction)
00163 {
00164    struct ast_datastore *datastore = NULL;
00165    struct pitchshift_data *shift = NULL;
00166 
00167 
00168    if (!f) {
00169       return 0;
00170    }
00171    if ((audiohook->status == AST_AUDIOHOOK_STATUS_DONE) ||
00172       (f->frametype != AST_FRAME_VOICE) ||
00173       ((f->subclass.codec != AST_FORMAT_SLINEAR) &&
00174       (f->subclass.codec != AST_FORMAT_SLINEAR16))) {
00175       return -1;
00176    }
00177 
00178    if (!(datastore = ast_channel_datastore_find(chan, &pitchshift_datastore, NULL))) {
00179       return -1;
00180    }
00181 
00182    shift = datastore->data;
00183 
00184    if (direction == AST_AUDIOHOOK_DIRECTION_WRITE) {
00185       pitch_shift(f, shift->tx.shift_amount, &shift->tx);
00186    } else {
00187       pitch_shift(f, shift->rx.shift_amount, &shift->rx);
00188    }
00189 
00190    return 0;
00191 }
00192 
00193 static int pitchshift_helper(struct ast_channel *chan, const char *cmd, char *data, const char *value)
00194 {
00195    struct ast_datastore *datastore = NULL;
00196    struct pitchshift_data *shift = NULL;
00197    int new = 0;
00198    float amount = 0;
00199 
00200    ast_channel_lock(chan);
00201    if (!(datastore = ast_channel_datastore_find(chan, &pitchshift_datastore, NULL))) {
00202       ast_channel_unlock(chan);
00203 
00204       if (!(datastore = ast_datastore_alloc(&pitchshift_datastore, NULL))) {
00205          return 0;
00206       }
00207       if (!(shift = ast_calloc(1, sizeof(*shift)))) {
00208          ast_datastore_free(datastore);
00209          return 0;
00210       }
00211 
00212       ast_audiohook_init(&shift->audiohook, AST_AUDIOHOOK_TYPE_MANIPULATE, "pitch_shift");
00213       shift->audiohook.manipulate_callback = pitchshift_cb;
00214       datastore->data = shift;
00215       new = 1;
00216    } else {
00217       ast_channel_unlock(chan);
00218       shift = datastore->data;
00219    }
00220 
00221 
00222    if (!strcasecmp(value, "highest")) {
00223       amount = HIGHEST;
00224    } else if (!strcasecmp(value, "higher")) {
00225       amount = HIGHER;
00226    } else if (!strcasecmp(value, "high")) {
00227       amount = HIGH;
00228    } else if (!strcasecmp(value, "lowest")) {
00229       amount = LOWEST;
00230    } else if (!strcasecmp(value, "lower")) {
00231       amount = LOWER;
00232    } else if (!strcasecmp(value, "low")) {
00233       amount = LOW;
00234    } else {
00235       if (!sscanf(value, "%30f", &amount) || (amount <= 0) || (amount > 4)) {
00236          goto cleanup_error;
00237       }
00238    }
00239 
00240    if (!strcasecmp(data, "rx")) {
00241       shift->rx.shift_amount = amount;
00242    } else if (!strcasecmp(data, "tx")) {
00243       shift->tx.shift_amount = amount;
00244    } else if (!strcasecmp(data, "both")) {
00245       shift->rx.shift_amount = amount;
00246       shift->tx.shift_amount = amount;
00247    } else {
00248       goto cleanup_error;
00249    }
00250 
00251    if (new) {
00252       ast_channel_lock(chan);
00253       ast_channel_datastore_add(chan, datastore);
00254       ast_channel_unlock(chan);
00255       ast_audiohook_attach(chan, &shift->audiohook);
00256    }
00257 
00258    return 0;
00259 
00260 cleanup_error:
00261 
00262    ast_log(LOG_ERROR, "Invalid argument provided to the %s function\n", cmd);
00263    if (new) {
00264       ast_datastore_free(datastore);
00265    }
00266    return -1;
00267 }
00268 
00269 static void smb_fft(float *fft_buffer, long fft_frame_size, long sign)
00270 {
00271    float wr, wi, arg, *p1, *p2, temp;
00272    float tr, ti, ur, ui, *p1r, *p1i, *p2r, *p2i;
00273    long i, bitm, j, le, le2, k;
00274 
00275    for (i = 2; i < 2 * fft_frame_size - 2; i += 2) {
00276       for (bitm = 2, j = 0; bitm < 2 * fft_frame_size; bitm <<= 1) {
00277          if (i & bitm) {
00278             j++;
00279          }
00280          j <<= 1;
00281       }
00282       if (i < j) {
00283          p1 = fft_buffer + i; p2 = fft_buffer + j;
00284          temp = *p1; *(p1++) = *p2;
00285          *(p2++) = temp; temp = *p1;
00286          *p1 = *p2; *p2 = temp;
00287       }
00288    }
00289    for (k = 0, le = 2; k < (long) (log(fft_frame_size) / log(2.) + .5); k++) {
00290       le <<= 1;
00291       le2 = le>>1;
00292       ur = 1.0;
00293       ui = 0.0;
00294       arg = M_PI / (le2>>1);
00295       wr = cos(arg);
00296       wi = sign * sin(arg);
00297       for (j = 0; j < le2; j += 2) {
00298          p1r = fft_buffer+j; p1i = p1r + 1;
00299          p2r = p1r + le2; p2i = p2r + 1;
00300          for (i = j; i < 2 * fft_frame_size; i += le) {
00301             tr = *p2r * ur - *p2i * ui;
00302             ti = *p2r * ui + *p2i * ur;
00303             *p2r = *p1r - tr; *p2i = *p1i - ti;
00304             *p1r += tr; *p1i += ti;
00305             p1r += le; p1i += le;
00306             p2r += le; p2i += le;
00307          }
00308          tr = ur * wr - ui * wi;
00309          ui = ur * wi + ui * wr;
00310          ur = tr;
00311       }
00312    }
00313 }
00314 
00315 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)
00316 {
00317    float *in_fifo = fft_data->in_fifo;
00318    float *out_fifo = fft_data->out_fifo;
00319    float *fft_worksp = fft_data->fft_worksp;
00320    float *last_phase = fft_data->last_phase;
00321    float *sum_phase = fft_data->sum_phase;
00322    float *output_accum = fft_data->output_accum;
00323    float *ana_freq = fft_data->ana_freq;
00324    float *ana_magn = fft_data->ana_magn;
00325    float *syn_freq = fft_data->syn_freq;
00326    float *sys_magn = fft_data->sys_magn;
00327 
00328    double magn, phase, tmp, window, real, imag;
00329    double freq_per_bin, expct;
00330    long i,k, qpd, index, in_fifo_latency, step_size, fft_frame_size2;
00331 
00332    /* set up some handy variables */
00333    fft_frame_size2 = fft_frame_size / 2;
00334    step_size = fft_frame_size / osamp;
00335    freq_per_bin = sample_rate / (double) fft_frame_size;
00336    expct = 2. * M_PI * (double) step_size / (double) fft_frame_size;
00337    in_fifo_latency = fft_frame_size-step_size;
00338 
00339    if (fft_data->gRover == 0) {
00340       fft_data->gRover = in_fifo_latency;
00341    }
00342 
00343    /* main processing loop */
00344    for (i = 0; i < num_samps_to_process; i++){
00345 
00346       /* As long as we have not yet collected enough data just read in */
00347       in_fifo[fft_data->gRover] = indata[i];
00348       outdata[i] = out_fifo[fft_data->gRover - in_fifo_latency];
00349       fft_data->gRover++;
00350 
00351       /* now we have enough data for processing */
00352       if (fft_data->gRover >= fft_frame_size) {
00353          fft_data->gRover = in_fifo_latency;
00354 
00355          /* do windowing and re,im interleave */
00356          for (k = 0; k < fft_frame_size;k++) {
00357             window = -.5 * cos(2. * M_PI * (double) k / (double) fft_frame_size) + .5;
00358             fft_worksp[2*k] = in_fifo[k] * window;
00359             fft_worksp[2*k+1] = 0.;
00360          }
00361 
00362          /* ***************** ANALYSIS ******************* */
00363          /* do transform */
00364          smb_fft(fft_worksp, fft_frame_size, -1);
00365 
00366          /* this is the analysis step */
00367          for (k = 0; k <= fft_frame_size2; k++) {
00368 
00369             /* de-interlace FFT buffer */
00370             real = fft_worksp[2*k];
00371             imag = fft_worksp[2*k+1];
00372 
00373             /* compute magnitude and phase */
00374             magn = 2. * sqrt(real * real + imag * imag);
00375             phase = atan2(imag, real);
00376 
00377             /* compute phase difference */
00378             tmp = phase - last_phase[k];
00379             last_phase[k] = phase;
00380 
00381             /* subtract expected phase difference */
00382             tmp -= (double) k * expct;
00383 
00384             /* map delta phase into +/- Pi interval */
00385             qpd = tmp / M_PI;
00386             if (qpd >= 0) {
00387                qpd += qpd & 1;
00388             } else {
00389                qpd -= qpd & 1;
00390             }
00391             tmp -= M_PI * (double) qpd;
00392 
00393             /* get deviation from bin frequency from the +/- Pi interval */
00394             tmp = osamp * tmp / (2. * M_PI);
00395 
00396             /* compute the k-th partials' true frequency */
00397             tmp = (double) k * freq_per_bin + tmp * freq_per_bin;
00398 
00399             /* store magnitude and true frequency in analysis arrays */
00400             ana_magn[k] = magn;
00401             ana_freq[k] = tmp;
00402 
00403          }
00404 
00405          /* ***************** PROCESSING ******************* */
00406          /* this does the actual pitch shifting */
00407          memset(sys_magn, 0, fft_frame_size * sizeof(float));
00408          memset(syn_freq, 0, fft_frame_size * sizeof(float));
00409          for (k = 0; k <= fft_frame_size2; k++) {
00410             index = k * pitchShift;
00411             if (index <= fft_frame_size2) {
00412                sys_magn[index] += ana_magn[k];
00413                syn_freq[index] = ana_freq[k] * pitchShift;
00414             }
00415          }
00416 
00417          /* ***************** SYNTHESIS ******************* */
00418          /* this is the synthesis step */
00419          for (k = 0; k <= fft_frame_size2; k++) {
00420 
00421             /* get magnitude and true frequency from synthesis arrays */
00422             magn = sys_magn[k];
00423             tmp = syn_freq[k];
00424 
00425             /* subtract bin mid frequency */
00426             tmp -= (double) k * freq_per_bin;
00427 
00428             /* get bin deviation from freq deviation */
00429             tmp /= freq_per_bin;
00430 
00431             /* take osamp into account */
00432             tmp = 2. * M_PI * tmp / osamp;
00433 
00434             /* add the overlap phase advance back in */
00435             tmp += (double) k * expct;
00436 
00437             /* accumulate delta phase to get bin phase */
00438             sum_phase[k] += tmp;
00439             phase = sum_phase[k];
00440 
00441             /* get real and imag part and re-interleave */
00442             fft_worksp[2*k] = magn * cos(phase);
00443             fft_worksp[2*k+1] = magn * sin(phase);
00444          }
00445 
00446          /* zero negative frequencies */
00447          for (k = fft_frame_size + 2; k < 2 * fft_frame_size; k++) {
00448             fft_worksp[k] = 0.;
00449          }
00450 
00451          /* do inverse transform */
00452          smb_fft(fft_worksp, fft_frame_size, 1);
00453 
00454          /* do windowing and add to output accumulator */
00455          for (k = 0; k < fft_frame_size; k++) {
00456             window = -.5 * cos(2. * M_PI * (double) k / (double) fft_frame_size) + .5;
00457             output_accum[k] += 2. * window * fft_worksp[2*k] / (fft_frame_size2 * osamp);
00458          }
00459          for (k = 0; k < step_size; k++) {
00460             out_fifo[k] = output_accum[k];
00461          }
00462 
00463          /* shift accumulator */
00464          memmove(output_accum, output_accum+step_size, fft_frame_size * sizeof(float));
00465 
00466          /* move input FIFO */
00467          for (k = 0; k < in_fifo_latency; k++) {
00468             in_fifo[k] = in_fifo[k+step_size];
00469          }
00470       }
00471    }
00472 }
00473 
00474 static int pitch_shift(struct ast_frame *f, float amount, struct fft_data *fft)
00475 {
00476    int16_t *fun = (int16_t *) f->data.ptr;
00477    int samples;
00478 
00479    /* an amount of 1 has no effect */
00480    if (!amount || amount == 1 || !fun || (f->samples % 32)) {
00481       return 0;
00482    }
00483    for (samples = 0; samples < f->samples; samples += 32) {
00484       smb_pitch_shift(amount, 32, MAX_FRAME_LENGTH, 32, ast_format_rate(f->subclass.codec), fun+samples, fun+samples, fft);
00485    }
00486 
00487    return 0;
00488 }
00489 
00490 static struct ast_custom_function pitch_shift_function = {
00491    .name = "PITCH_SHIFT",
00492    .write = pitchshift_helper,
00493 };
00494 
00495 static int unload_module(void)
00496 {
00497    return ast_custom_function_unregister(&pitch_shift_function);
00498 }
00499 
00500 static int load_module(void)
00501 {
00502    int res = ast_custom_function_register(&pitch_shift_function);
00503    return res ? AST_MODULE_LOAD_DECLINE : AST_MODULE_LOAD_SUCCESS;
00504 }
00505 
00506 AST_MODULE_INFO_STANDARD(ASTERISK_GPL_KEY, "Audio Effects Dialplan Functions");

Generated on Wed Apr 6 11:29:45 2011 for Asterisk - The Open Source Telephony Project by  doxygen 1.4.7