diff options
| -rw-r--r-- | LICENSE | 4 | ||||
| -rw-r--r-- | src/Makefile.am | 5 | ||||
| -rw-r--r-- | src/modules/echo-cancel/adrian-aec.c | 233 | ||||
| -rw-r--r-- | src/modules/echo-cancel/adrian-aec.h | 370 | ||||
| -rw-r--r-- | src/modules/echo-cancel/adrian-license.txt | 17 | ||||
| -rw-r--r-- | src/modules/echo-cancel/adrian.c | 121 | ||||
| -rw-r--r-- | src/modules/echo-cancel/adrian.h | 31 | ||||
| -rw-r--r-- | src/modules/echo-cancel/echo-cancel.h | 14 | ||||
| -rw-r--r-- | src/modules/echo-cancel/module-echo-cancel.c | 8 |
9 files changed, 802 insertions, 1 deletions
@@ -10,4 +10,8 @@ LGPL licensed and the server part ('libpulsecore') as being GPL licensed. Since the PulseAudio daemon and the modules link to 'libpulsecore' they are of course also GPL licensed. +Andre Adrian's echo cancellation implementation is licensed under a less +restrictive license - see src/modules/echo-cancel/adrian-license.txt for +details. + -- Lennart Poettering, April 20th, 2006. diff --git a/src/Makefile.am b/src/Makefile.am index 242532ca..09804886 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -1605,7 +1605,10 @@ module_suspend_on_idle_la_LIBADD = $(AM_LIBADD) libpulsecore-@PA_MAJORMINORMICRO module_suspend_on_idle_la_CFLAGS = $(AM_CFLAGS) # echo-cancel module -module_echo_cancel_la_SOURCES = modules/echo-cancel/module-echo-cancel.c modules/echo-cancel/speex.c +module_echo_cancel_la_SOURCES = modules/echo-cancel/module-echo-cancel.c \ + modules/echo-cancel/speex.c \ + modules/echo-cancel/adrian-aec.c \ + modules/echo-cancel/adrian.c module_echo_cancel_la_LDFLAGS = $(MODULE_LDFLAGS) module_echo_cancel_la_LIBADD = $(AM_LIBADD) libpulsecore-@PA_MAJORMINORMICRO@.la libpulsecommon-@PA_MAJORMINORMICRO@.la libpulse.la $(LIBSPEEX_LIBS) module_echo_cancel_la_CFLAGS = $(AM_CFLAGS) $(LIBSPEEX_CFLAGS) diff --git a/src/modules/echo-cancel/adrian-aec.c b/src/modules/echo-cancel/adrian-aec.c new file mode 100644 index 00000000..69107c75 --- /dev/null +++ b/src/modules/echo-cancel/adrian-aec.c @@ -0,0 +1,233 @@ +/* aec.cpp + * + * Copyright (C) DFS Deutsche Flugsicherung (2004, 2005). + * All Rights Reserved. + * + * Acoustic Echo Cancellation NLMS-pw algorithm + * + * Version 0.3 filter created with www.dsptutor.freeuk.com + * Version 0.3.1 Allow change of stability parameter delta + * Version 0.4 Leaky Normalized LMS - pre whitening algorithm + */ + +#include <math.h> +#include <string.h> + +#include <pulse/xmalloc.h> + +#include "adrian-aec.h" + +/* Vector Dot Product */ +static REAL dotp(REAL a[], REAL b[]) +{ + REAL sum0 = 0.0, sum1 = 0.0; + int j; + + for (j = 0; j < NLMS_LEN; j += 2) { + // optimize: partial loop unrolling + sum0 += a[j] * b[j]; + sum1 += a[j + 1] * b[j + 1]; + } + return sum0 + sum1; +} + + +AEC* AEC_init(int RATE) +{ + AEC *a = pa_xnew(AEC, 1); + a->hangover = 0; + memset(a->x, 0, sizeof(a->x)); + memset(a->xf, 0, sizeof(a->xf)); + memset(a->w, 0, sizeof(a->w)); + a->j = NLMS_EXT; + a->delta = 0.0f; + AEC_setambient(a, NoiseFloor); + a->dfast = a->dslow = M75dB_PCM; + a->xfast = a->xslow = M80dB_PCM; + a->gain = 1.0f; + a->Fx = IIR1_init(2000.0f/RATE); + a->Fe = IIR1_init(2000.0f/RATE); + a->cutoff = FIR_HP_300Hz_init(); + a->acMic = IIR_HP_init(); + a->acSpk = IIR_HP_init(); + + a->aes_y2 = M0dB; + + a->fdwdisplay = -1; + a->dumpcnt = 0; + memset(a->ws, 0, sizeof(a->ws)); + + return a; +} + +// Adrian soft decision DTD +// (Dual Average Near-End to Far-End signal Ratio DTD) +// This algorithm uses exponential smoothing with differnt +// ageing parameters to get fast and slow near-end and far-end +// signal averages. The ratio of NFRs term +// (dfast / xfast) / (dslow / xslow) is used to compute the stepsize +// A ratio value of 2.5 is mapped to stepsize 0, a ratio of 0 is +// mapped to 1.0 with a limited linear function. +static float AEC_dtd(AEC *a, REAL d, REAL x) +{ + float stepsize; + float ratio, M; + + // fast near-end and far-end average + a->dfast += ALPHAFAST * (fabsf(d) - a->dfast); + a->xfast += ALPHAFAST * (fabsf(x) - a->xfast); + + // slow near-end and far-end average + a->dslow += ALPHASLOW * (fabsf(d) - a->dslow); + a->xslow += ALPHASLOW * (fabsf(x) - a->xslow); + + if (a->xfast < M70dB_PCM) { + return 0.0; // no Spk signal + } + + if (a->dfast < M70dB_PCM) { + return 0.0; // no Mic signal + } + + // ratio of NFRs + ratio = (a->dfast * a->xslow) / (a->dslow * a->xfast); + + // begrenzte lineare Kennlinie + M = (STEPY2 - STEPY1) / (STEPX2 - STEPX1); + if (ratio < STEPX1) { + stepsize = STEPY1; + } else if (ratio > STEPX2) { + stepsize = STEPY2; + } else { + // Punktrichtungsform einer Geraden + stepsize = M * (ratio - STEPX1) + STEPY1; + } + + return stepsize; +} + + +static void AEC_leaky(AEC *a) +// The xfast signal is used to charge the hangover timer to Thold. +// When hangover expires (no Spk signal for some time) the vector w +// is erased. This is my implementation of Leaky NLMS. +{ + if (a->xfast >= M70dB_PCM) { + // vector w is valid for hangover Thold time + a->hangover = Thold; + } else { + if (a->hangover > 1) { + --(a->hangover); + } else if (1 == a->hangover) { + --(a->hangover); + // My Leaky NLMS is to erase vector w when hangover expires + memset(a->w, 0, sizeof(a->w)); + } + } +} + + +#if 0 +void AEC::openwdisplay() { + // open TCP connection to program wdisplay.tcl + fdwdisplay = socket_async("127.0.0.1", 50999); +}; +#endif + + +static REAL AEC_nlms_pw(AEC *a, REAL d, REAL x_, float stepsize) +{ + REAL e; + REAL ef; + a->x[a->j] = x_; + a->xf[a->j] = IIR1_highpass(a->Fx, x_); // pre-whitening of x + + // calculate error value + // (mic signal - estimated mic signal from spk signal) + e = d; + if (a->hangover > 0) { + e -= dotp(a->w, a->x + a->j); + } + ef = IIR1_highpass(a->Fe, e); // pre-whitening of e + + // optimize: iterative dotp(xf, xf) + a->dotp_xf_xf += (a->xf[a->j] * a->xf[a->j] - a->xf[a->j + NLMS_LEN - 1] * a->xf[a->j + NLMS_LEN - 1]); + + if (stepsize > 0.0) { + // calculate variable step size + REAL mikro_ef = stepsize * ef / a->dotp_xf_xf; + + // update tap weights (filter learning) + int i; + for (i = 0; i < NLMS_LEN; i += 2) { + // optimize: partial loop unrolling + a->w[i] += mikro_ef * a->xf[i + a->j]; + a->w[i + 1] += mikro_ef * a->xf[i + a->j + 1]; + } + } + + if (--(a->j) < 0) { + // optimize: decrease number of memory copies + a->j = NLMS_EXT; + memmove(a->x + a->j + 1, a->x, (NLMS_LEN - 1) * sizeof(REAL)); + memmove(a->xf + a->j + 1, a->xf, (NLMS_LEN - 1) * sizeof(REAL)); + } + + // Saturation + if (e > MAXPCM) { + return MAXPCM; + } else if (e < -MAXPCM) { + return -MAXPCM; + } else { + return e; + } +} + + +int AEC_doAEC(AEC *a, int d_, int x_) +{ + REAL d = (REAL) d_; + REAL x = (REAL) x_; + + // Mic Highpass Filter - to remove DC + d = IIR_HP_highpass(a->acMic, d); + + // Mic Highpass Filter - cut-off below 300Hz + d = FIR_HP_300Hz_highpass(a->cutoff, d); + + // Amplify, for e.g. Soundcards with -6dB max. volume + d *= a->gain; + + // Spk Highpass Filter - to remove DC + x = IIR_HP_highpass(a->acSpk, x); + + // Double Talk Detector + a->stepsize = AEC_dtd(a, d, x); + + // Leaky (ageing of vector w) + AEC_leaky(a); + + // Acoustic Echo Cancellation + d = AEC_nlms_pw(a, d, x, a->stepsize); + +#if 0 + if (fdwdisplay >= 0) { + if (++dumpcnt >= (WIDEB*RATE/10)) { + // wdisplay creates 10 dumps per seconds = large CPU load! + dumpcnt = 0; + write(fdwdisplay, ws, DUMP_LEN*sizeof(float)); + // we don't check return value. This is not production quality!!! + memset(ws, 0, sizeof(ws)); + } else { + int i; + for (i = 0; i < DUMP_LEN; i += 2) { + // optimize: partial loop unrolling + ws[i] += w[i]; + ws[i + 1] += w[i + 1]; + } + } + } +#endif + + return (int) d; +} diff --git a/src/modules/echo-cancel/adrian-aec.h b/src/modules/echo-cancel/adrian-aec.h new file mode 100644 index 00000000..1f5b090a --- /dev/null +++ b/src/modules/echo-cancel/adrian-aec.h @@ -0,0 +1,370 @@ +/* aec.h + * + * Copyright (C) DFS Deutsche Flugsicherung (2004, 2005). + * All Rights Reserved. + * Author: Andre Adrian + * + * Acoustic Echo Cancellation Leaky NLMS-pw algorithm + * + * Version 0.3 filter created with www.dsptutor.freeuk.com + * Version 0.3.1 Allow change of stability parameter delta + * Version 0.4 Leaky Normalized LMS - pre whitening algorithm + */ + +#ifndef _AEC_H /* include only once */ + +#define WIDEB 2 + +// use double if your CPU does software-emulation of float +typedef float REAL; + +/* dB Values */ +#define M0dB 1.0f +#define M3dB 0.71f +#define M6dB 0.50f +#define M9dB 0.35f +#define M12dB 0.25f +#define M18dB 0.125f +#define M24dB 0.063f + +/* dB values for 16bit PCM */ +/* MxdB_PCM = 32767 * 10 ^(x / 20) */ +#define M10dB_PCM 10362.0f +#define M20dB_PCM 3277.0f +#define M25dB_PCM 1843.0f +#define M30dB_PCM 1026.0f +#define M35dB_PCM 583.0f +#define M40dB_PCM 328.0f +#define M45dB_PCM 184.0f +#define M50dB_PCM 104.0f +#define M55dB_PCM 58.0f +#define M60dB_PCM 33.0f +#define M65dB_PCM 18.0f +#define M70dB_PCM 10.0f +#define M75dB_PCM 6.0f +#define M80dB_PCM 3.0f +#define M85dB_PCM 2.0f +#define M90dB_PCM 1.0f + +#define MAXPCM 32767.0f + +/* Design constants (Change to fine tune the algorithms */ + +/* The following values are for hardware AEC and studio quality + * microphone */ + +/* NLMS filter length in taps (samples). A longer filter length gives + * better Echo Cancellation, but maybe slower convergence speed and + * needs more CPU power (Order of NLMS is linear) */ +#define NLMS_LEN (100*WIDEB*8) + +/* Vector w visualization length in taps (samples). + * Must match argv value for wdisplay.tcl */ +#define DUMP_LEN (40*WIDEB*8) + +/* minimum energy in xf. Range: M70dB_PCM to M50dB_PCM. Should be equal + * to microphone ambient Noise level */ +#define NoiseFloor M55dB_PCM + +/* Leaky hangover in taps. + */ +#define Thold (60 * WIDEB * 8) + +// Adrian soft decision DTD +// left point. X is ratio, Y is stepsize +#define STEPX1 1.0 +#define STEPY1 1.0 +// right point. STEPX2=2.0 is good double talk, 3.0 is good single talk. +#define STEPX2 2.5 +#define STEPY2 0 +#define ALPHAFAST (1.0f / 100.0f) +#define ALPHASLOW (1.0f / 20000.0f) + + + +/* Ageing multiplier for LMS memory vector w */ +#define Leaky 0.9999f + +/* Double Talk Detector Speaker/Microphone Threshold. Range <=1 + * Large value (M0dB) is good for Single-Talk Echo cancellation, + * small value (M12dB) is good for Doulbe-Talk AEC */ +#define GeigelThreshold M6dB + +/* for Non Linear Processor. Range >0 to 1. Large value (M0dB) is good + * for Double-Talk, small value (M12dB) is good for Single-Talk */ +#define NLPAttenuation M12dB + +/* Below this line there are no more design constants */ + +typedef struct IIR_HP IIR_HP; + +/* Exponential Smoothing or IIR Infinite Impulse Response Filter */ +struct IIR_HP { + REAL x; +}; + +static IIR_HP* IIR_HP_init(void) { + IIR_HP *i = pa_xnew(IIR_HP, 1); + i->x = 0.0f; + return i; + } + +static REAL IIR_HP_highpass(IIR_HP *i, REAL in) { + const REAL a0 = 0.01f; /* controls Transfer Frequency */ + /* Highpass = Signal - Lowpass. Lowpass = Exponential Smoothing */ + i->x += a0 * (in - i->x); + return in - i->x; + }; + +typedef struct FIR_HP_300Hz FIR_HP_300Hz; + +#if WIDEB==1 +/* 17 taps FIR Finite Impulse Response filter + * Coefficients calculated with + * www.dsptutor.freeuk.com/KaiserFilterDesign/KaiserFilterDesign.html + */ +class FIR_HP_300Hz { + REAL z[18]; + +public: + FIR_HP_300Hz() { + memset(this, 0, sizeof(FIR_HP_300Hz)); + } + + REAL highpass(REAL in) { + const REAL a[18] = { + // Kaiser Window FIR Filter, Filter type: High pass + // Passband: 300.0 - 4000.0 Hz, Order: 16 + // Transition band: 75.0 Hz, Stopband attenuation: 10.0 dB + -0.034870606, -0.039650206, -0.044063766, -0.04800318, + -0.051370874, -0.054082647, -0.056070227, -0.057283327, + 0.8214126, -0.057283327, -0.056070227, -0.054082647, + -0.051370874, -0.04800318, -0.044063766, -0.039650206, + -0.034870606, 0.0 + }; + memmove(z + 1, z, 17 * sizeof(REAL)); + z[0] = in; + REAL sum0 = 0.0, sum1 = 0.0; + int j; + + for (j = 0; j < 18; j += 2) { + // optimize: partial loop unrolling + sum0 += a[j] * z[j]; + sum1 += a[j + 1] * z[j + 1]; + } + return sum0 + sum1; + } +}; + +#else + +/* 35 taps FIR Finite Impulse Response filter + * Passband 150Hz to 4kHz for 8kHz sample rate, 300Hz to 8kHz for 16kHz + * sample rate. + * Coefficients calculated with + * www.dsptutor.freeuk.com/KaiserFilterDesign/KaiserFilterDesign.html + */ +struct FIR_HP_300Hz { + REAL z[36]; +}; + +static FIR_HP_300Hz* FIR_HP_300Hz_init(void) { + FIR_HP_300Hz *ret = pa_xnew(FIR_HP_300Hz, 1); + memset(ret, 0, sizeof(FIR_HP_300Hz)); + return ret; + } + +static REAL FIR_HP_300Hz_highpass(FIR_HP_300Hz *f, REAL in) { + REAL sum0 = 0.0, sum1 = 0.0; + int j; + const REAL a[36] = { + // Kaiser Window FIR Filter, Filter type: High pass + // Passband: 150.0 - 4000.0 Hz, Order: 34 + // Transition band: 34.0 Hz, Stopband attenuation: 10.0 dB + -0.016165324, -0.017454365, -0.01871232, -0.019931411, + -0.021104068, -0.022222936, -0.02328091, -0.024271343, + -0.025187887, -0.02602462, -0.026776174, -0.027437767, + -0.028004972, -0.028474221, -0.028842418, -0.029107114, + -0.02926664, 0.8524841, -0.02926664, -0.029107114, + -0.028842418, -0.028474221, -0.028004972, -0.027437767, + -0.026776174, -0.02602462, -0.025187887, -0.024271343, + -0.02328091, -0.022222936, -0.021104068, -0.019931411, + -0.01871232, -0.017454365, -0.016165324, 0.0 + }; + memmove(f->z + 1, f->z, 35 * sizeof(REAL)); + f->z[0] = in; + + for (j = 0; j < 36; j += 2) { + // optimize: partial loop unrolling + sum0 += a[j] * f->z[j]; + sum1 += a[j + 1] * f->z[j + 1]; + } + return sum0 + sum1; + } +#endif + +typedef struct IIR1 IIR1; + +/* Recursive single pole IIR Infinite Impulse response High-pass filter + * + * Reference: The Scientist and Engineer's Guide to Digital Processing + * + * output[N] = A0 * input[N] + A1 * input[N-1] + B1 * output[N-1] + * + * X = exp(-2.0 * pi * Fc) + * A0 = (1 + X) / 2 + * A1 = -(1 + X) / 2 + * B1 = X + * Fc = cutoff freq / sample rate + */ +struct IIR1 { + REAL in0, out0; + REAL a0, a1, b1; +}; + +#if 0 + IIR1() { + memset(this, 0, sizeof(IIR1)); + } +#endif + +static IIR1* IIR1_init(REAL Fc) { + IIR1 *i = pa_xnew(IIR1, 1); + i->b1 = expf(-2.0f * M_PI * Fc); + i->a0 = (1.0f + i->b1) / 2.0f; + i->a1 = -(i->a0); + i->in0 = 0.0f; + i->out0 = 0.0f; + return i; + } + +static REAL IIR1_highpass(IIR1 *i, REAL in) { + REAL out = i->a0 * in + i->a1 * i->in0 + i->b1 * i->out0; + i->in0 = in; + i->out0 = out; + return out; + } + + +#if 0 +/* Recursive two pole IIR Infinite Impulse Response filter + * Coefficients calculated with + * http://www.dsptutor.freeuk.com/IIRFilterDesign/IIRFiltDes102.html + */ +class IIR2 { + REAL x[2], y[2]; + +public: + IIR2() { + memset(this, 0, sizeof(IIR2)); + } + + REAL highpass(REAL in) { + // Butterworth IIR filter, Filter type: HP + // Passband: 2000 - 4000.0 Hz, Order: 2 + const REAL a[] = { 0.29289323f, -0.58578646f, 0.29289323f }; + const REAL b[] = { 1.3007072E-16f, 0.17157288f }; + REAL out = + a[0] * in + a[1] * x[0] + a[2] * x[1] - b[0] * y[0] - b[1] * y[1]; + + x[1] = x[0]; + x[0] = in; + y[1] = y[0]; + y[0] = out; + return out; + } +}; +#endif + + +// Extention in taps to reduce mem copies +#define NLMS_EXT (10*8) + +// block size in taps to optimize DTD calculation +#define DTD_LEN 16 + +typedef struct AEC AEC; + +struct AEC { + // Time domain Filters + IIR_HP *acMic, *acSpk; // DC-level remove Highpass) + FIR_HP_300Hz *cutoff; // 150Hz cut-off Highpass + REAL gain; // Mic signal amplify + IIR1 *Fx, *Fe; // pre-whitening Highpass for x, e + + // Adrian soft decision DTD (Double Talk Detector) + REAL dfast, xfast; + REAL dslow, xslow; + + // NLMS-pw + REAL x[NLMS_LEN + NLMS_EXT]; // tap delayed loudspeaker signal + REAL xf[NLMS_LEN + NLMS_EXT]; // pre-whitening tap delayed signal + REAL w[NLMS_LEN]; // tap weights + int j; // optimize: less memory copies + double dotp_xf_xf; // double to avoid loss of precision + float delta; // noise floor to stabilize NLMS + + // AES + float aes_y2; // not in use! + + // w vector visualization + REAL ws[DUMP_LEN]; // tap weights sums + int fdwdisplay; // TCP file descriptor + int dumpcnt; // wdisplay output counter + + // variables are public for visualization + int hangover; + float stepsize; +}; + +/* Double-Talk Detector + * + * in d: microphone sample (PCM as REALing point value) + * in x: loudspeaker sample (PCM as REALing point value) + * return: from 0 for doubletalk to 1.0 for single talk + */ +static float AEC_dtd(AEC *a, REAL d, REAL x); + +static void AEC_leaky(AEC *a); + +/* Normalized Least Mean Square Algorithm pre-whitening (NLMS-pw) + * The LMS algorithm was developed by Bernard Widrow + * book: Haykin, Adaptive Filter Theory, 4. edition, Prentice Hall, 2002 + * + * in d: microphone sample (16bit PCM value) + * in x_: loudspeaker sample (16bit PCM value) + * in stepsize: NLMS adaptation variable + * return: echo cancelled microphone sample + */ +static REAL AEC_nlms_pw(AEC *a, REAL d, REAL x_, float stepsize); + + AEC* AEC_init(int RATE); + +/* Acoustic Echo Cancellation and Suppression of one sample + * in d: microphone signal with echo + * in x: loudspeaker signal + * return: echo cancelled microphone signal + */ + int AEC_doAEC(AEC *a, int d_, int x_); + +static float AEC_getambient(AEC *a) { + return a->dfast; + }; +static void AEC_setambient(AEC *a, float Min_xf) { + a->dotp_xf_xf -= a->delta; // subtract old delta + a->delta = (NLMS_LEN-1) * Min_xf * Min_xf; + a->dotp_xf_xf += a->delta; // add new delta + }; +static void AEC_setgain(AEC *a, float gain_) { + a->gain = gain_; + }; +#if 0 + void AEC_openwdisplay(AEC *a); +#endif +static void AEC_setaes(AEC *a, float aes_y2_) { + a->aes_y2 = aes_y2_; + }; +static double AEC_max_dotp_xf_xf(AEC *a, double u); + +#define _AEC_H +#endif diff --git a/src/modules/echo-cancel/adrian-license.txt b/src/modules/echo-cancel/adrian-license.txt new file mode 100644 index 00000000..7c06efd0 --- /dev/null +++ b/src/modules/echo-cancel/adrian-license.txt @@ -0,0 +1,17 @@ + Copyright (C) DFS Deutsche Flugsicherung (2004). All Rights Reserved. + + You are allowed to use this source code in any open source or closed + source software you want. You are allowed to use the algorithms for a + hardware solution. You are allowed to modify the source code. + You are not allowed to remove the name of the author from this memo or + from the source code files. You are not allowed to monopolize the + source code or the algorithms behind the source code as your + intellectual property. This source code is free of royalty and comes + with no warranty. + +--- The following does not apply to the PulseAudio module --- + + Please see g711/gen-lic.txt for the ITU-T G.711 codec copyright. + Please see gsm/gen-lic.txt for the ITU-T GSM codec copyright. + Please see ilbc/COPYRIGHT and ilbc/NOTICE for the IETF iLBC codec + copyright. diff --git a/src/modules/echo-cancel/adrian.c b/src/modules/echo-cancel/adrian.c new file mode 100644 index 00000000..86c22cb3 --- /dev/null +++ b/src/modules/echo-cancel/adrian.c @@ -0,0 +1,121 @@ +/*** + This file is part of PulseAudio. + + Copyright 2010 Arun Raghavan <arun.raghavan@collabora.co.uk> + + Contributor: Wim Taymans <wim.taymans@gmail.com> + + The actual implementation is taken from the sources at + http://andreadrian.de/intercom/ - for the license, look for + adrian-license.txt in the same directory as this file. + + PulseAudio is free software; you can redistribute it and/or modify + it under the terms of the GNU Lesser General Public License as published + by the Free Software Foundation; either version 2.1 of the License, + or (at your option) any later version. + + PulseAudio is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + You should have received a copy of the GNU Lesser General Public License + along with PulseAudio; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 + USA. +***/ + +#ifdef HAVE_CONFIG_H +#include <config.h> +#endif + +#include <pulsecore/modargs.h> +#include "echo-cancel.h" + +/* should be between 10-20 ms */ +#define DEFAULT_FRAME_SIZE_MS 20 + +static const char* const valid_modargs[] = { + "frame_size_ms", + NULL +}; + +static void pa_adrian_ec_fixate_spec(pa_sample_spec *source_ss, pa_channel_map *source_map, + pa_sample_spec *sink_ss, pa_channel_map *sink_map) +{ + source_ss->format = PA_SAMPLE_S16LE; + source_ss->channels = 1; + pa_channel_map_init_mono(source_map); + + *sink_ss = *source_ss; + *sink_map = *source_map; +} + +pa_bool_t pa_adrian_ec_init(pa_echo_canceller *ec, + pa_sample_spec *source_ss, pa_channel_map *source_map, + pa_sample_spec *sink_ss, pa_channel_map *sink_map, + const char *args) +{ + int framelen, rate; + uint32_t frame_size_ms; + pa_modargs *ma; + + if (!(ma = pa_modargs_new(args, valid_modargs))) { + pa_log("Failed to parse submodule arguments."); + goto fail; + } + + frame_size_ms = DEFAULT_FRAME_SIZE_MS; + if (pa_modargs_get_value_u32(ma, "frame_size_ms", &frame_size_ms) < 0 || frame_size_ms < 1 || frame_size_ms > 200) { + pa_log("Invalid frame_size_ms specification"); + goto fail; + } + + pa_adrian_ec_fixate_spec(source_ss, source_map, sink_ss, sink_map); + + rate = source_ss->rate; + framelen = (rate * frame_size_ms) / 1000; + + ec->params.priv.adrian.blocksize = framelen * pa_frame_size (source_ss); + + pa_log_debug ("Using framelen %d, blocksize %lld, channels %d, rate %d", framelen, (long long) ec->params.priv.adrian.blocksize, source_ss->channels, source_ss->rate); + + ec->params.priv.adrian.aec = AEC_init(rate); + if (!ec->params.priv.adrian.aec) + goto fail; + + pa_modargs_free(ma); + return TRUE; + +fail: + if (ma) + pa_modargs_free(ma); + return FALSE; +} + +void pa_adrian_ec_run(pa_echo_canceller *ec, uint8_t *rec, uint8_t *play, uint8_t *out) +{ + unsigned int i; + + for (i = 0; i < ec->params.priv.adrian.blocksize; i += 2) { + /* We know it's S16LE mono data */ + int r = (((int8_t) rec[i + 1]) << 8) | rec[i]; + int p = (((int8_t) play[i + 1]) << 8) | play[i]; + int res; + + res = AEC_doAEC(ec->params.priv.adrian.aec, r, p); + out[i] = (uint8_t) (res & 0xff); + out[i + 1] = (uint8_t) ((res >> 8) & 0xff); + } +} + +void pa_adrian_ec_done(pa_echo_canceller *ec) +{ + pa_xfree(ec->params.priv.adrian.aec); + ec->params.priv.adrian.aec = NULL; +} + +uint32_t pa_adrian_ec_get_block_size(pa_echo_canceller *ec) +{ + return ec->params.priv.adrian.blocksize; +} diff --git a/src/modules/echo-cancel/adrian.h b/src/modules/echo-cancel/adrian.h new file mode 100644 index 00000000..d02e934d --- /dev/null +++ b/src/modules/echo-cancel/adrian.h @@ -0,0 +1,31 @@ +/*** + This file is part of PulseAudio. + + Copyright 2010 Arun Raghavan <arun.raghavan@collabora.co.uk> + + The actual implementation is taken from the sources at + http://andreadrian.de/intercom/ - for the license, look for + adrian-license.txt in the same directory as this file. + + PulseAudio is free software; you can redistribute it and/or modify + it under the terms of the GNU Lesser General Public License as published + by the Free Software Foundation; either version 2.1 of the License, + or (at your option) any later version. + + PulseAudio is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + You should have received a copy of the GNU Lesser General Public License + along with PulseAudio; if not, write to the Free Software + Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 + USA. +***/ + +/* Forward declarations */ + +typedef struct AEC AEC; + +AEC* AEC_init(int RATE); +int AEC_doAEC(AEC *a, int d_, int x_); diff --git a/src/modules/echo-cancel/echo-cancel.h b/src/modules/echo-cancel/echo-cancel.h index 205c4d14..65e0e240 100644 --- a/src/modules/echo-cancel/echo-cancel.h +++ b/src/modules/echo-cancel/echo-cancel.h @@ -28,6 +28,7 @@ #include <pulsecore/macro.h> #include <speex/speex_echo.h> +#include "adrian.h" /* Common data structures */ @@ -39,6 +40,10 @@ struct pa_echo_canceller_params { uint32_t blocksize; SpeexEchoState *state; } speex; + struct { + uint32_t blocksize; + AEC *aec; + } adrian; /* each canceller-specific structure goes here */ } priv; }; @@ -67,3 +72,12 @@ pa_bool_t pa_speex_ec_init(pa_echo_canceller *ec, void pa_speex_ec_run(pa_echo_canceller *ec, uint8_t *rec, uint8_t *play, uint8_t *out); void pa_speex_ec_done(pa_echo_canceller *ec); uint32_t pa_speex_ec_get_block_size(pa_echo_canceller *ec); + +/* Adrian Andre's echo canceller */ +pa_bool_t pa_adrian_ec_init(pa_echo_canceller *ec, + pa_sample_spec *source_ss, pa_channel_map *source_map, + pa_sample_spec *sink_ss, pa_channel_map *sink_map, + const char *args); +void pa_adrian_ec_run(pa_echo_canceller *ec, uint8_t *rec, uint8_t *play, uint8_t *out); +void pa_adrian_ec_done(pa_echo_canceller *ec); +uint32_t pa_adrian_ec_get_block_size(pa_echo_canceller *ec); diff --git a/src/modules/echo-cancel/module-echo-cancel.c b/src/modules/echo-cancel/module-echo-cancel.c index 6a88167b..75f74d34 100644 --- a/src/modules/echo-cancel/module-echo-cancel.c +++ b/src/modules/echo-cancel/module-echo-cancel.c @@ -82,6 +82,7 @@ PA_MODULE_USAGE( /* NOTE: Make sure the enum and ec_table are maintained in the correct order */ enum { PA_ECHO_CANCELLER_SPEEX, + PA_ECHO_CANCELLER_ADRIAN, }; #define DEFAULT_ECHO_CANCELLER PA_ECHO_CANCELLER_SPEEX @@ -94,6 +95,13 @@ static const pa_echo_canceller ec_table[] = { .done = pa_speex_ec_done, .get_block_size = pa_speex_ec_get_block_size, }, + { + /* Adrian Andre's NLMS implementation */ + .init = pa_adrian_ec_init, + .run = pa_adrian_ec_run, + .done = pa_adrian_ec_done, + .get_block_size = pa_adrian_ec_get_block_size, + }, }; #define DEFAULT_ADJUST_TIME_USEC (1*PA_USEC_PER_SEC) |