You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
sems/core/AmPlayoutBuffer.cpp

574 lines
14 KiB

/*
* Copyright (C) 2005-2006 iptelorg GmbH
*
* This file is part of SEMS, a free SIP media server.
*
* SEMS is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version. This program is released under
* the GPL with the additional exemption that compiling, linking,
* and/or using OpenSSL is allowed.
*
* For a license to use the SEMS software under conditions
* other than those described here, or to purchase support for this
* software, please contact iptel.org by e-mail at the following addresses:
* info@iptel.org
*
* SEMS 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 General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "AmPlayoutBuffer.h"
#include "AmAudio.h"
#include "AmRtpAudio.h"
#define SEARCH_OFFSET 140
#define SEARCH_REGION 110
#define DELTA 5
#define TSM_MAX_SCALE 2.0
#define TSM_MIN_SCALE 0.5
// only scale if 0.9 < f < 1.1
#define SCALE_FACTOR_START 0.1
#define PI 3.14
#define MAX_DELAY sample_rate*1 /* 1 second */
AmPlayoutBuffer::AmPlayoutBuffer(AmPLCBuffer *plcbuffer, unsigned int sample_rate)
: r_ts(0),w_ts(0), sample_rate(sample_rate),
last_ts_i(false), recv_offset_i(false),
m_plcbuffer(plcbuffer)
{
buffer.clear_all();
}
void AmPlayoutBuffer::direct_write_buffer(unsigned int ts, ShortSample* buf, unsigned int len)
{
buffer_put(w_ts,buf,len);
}
void AmPlayoutBuffer::write(u_int32_t ref_ts, u_int32_t rtp_ts,
int16_t* buf, u_int32_t len, bool begin_talk)
{
unsigned int mapped_ts;
if(!recv_offset_i)
{
recv_offset = rtp_ts - ref_ts;
recv_offset_i = true;
DBG("initialized recv_offset with %u (%u - %u)\n",
recv_offset, ref_ts, rtp_ts);
mapped_ts = r_ts = w_ts = ref_ts;
}
else {
mapped_ts = rtp_ts - recv_offset;
// resync
if( ts_less()(mapped_ts, ref_ts - MAX_DELAY/2) ||
!ts_less()(mapped_ts, ref_ts + MAX_DELAY) ){
DBG("resync needed: reference ts = %u; write ts = %u\n",
ref_ts, mapped_ts);
recv_offset = rtp_ts - ref_ts;
mapped_ts = r_ts = w_ts = ref_ts;
}
}
if(!last_ts_i)
{
last_ts = mapped_ts;
last_ts_i = true;
}
if(ts_less()(last_ts, mapped_ts) && !begin_talk
&& (mapped_ts - last_ts <= PLC_MAX_SAMPLES))
{
unsigned char tmp[AUDIO_BUFFER_SIZE * 2];
int l_size = m_plcbuffer->conceal_loss(mapped_ts - last_ts, tmp);
if (l_size>0)
{
direct_write_buffer(last_ts, (ShortSample*)tmp, PCM16_B2S(l_size));
}
}
m_plcbuffer->add_to_history(buf, PCM16_S2B(len));
write_buffer(ref_ts, mapped_ts, buf, len);
// update last_ts to end of received packet
// if not out-of-sequence
if (ts_less()(last_ts, mapped_ts) || last_ts == mapped_ts)
last_ts = mapped_ts + len;
}
void AmPlayoutBuffer::write_buffer(u_int32_t ref_ts, u_int32_t ts, int16_t* buf, u_int32_t len)
{
buffer_put(w_ts,buf,len);
}
u_int32_t AmPlayoutBuffer::read(u_int32_t ts, int16_t* buf, u_int32_t len)
{
if(ts_less()(r_ts,w_ts)){
u_int32_t rlen=0;
if(ts_less()(r_ts+PCM16_B2S(AUDIO_BUFFER_SIZE),w_ts))
rlen = PCM16_B2S(AUDIO_BUFFER_SIZE);
else
rlen = w_ts - r_ts;
buffer_get(r_ts,buf,rlen);
return rlen;
}
return 0;
}
void AmPlayoutBuffer::buffer_put(unsigned int ts, ShortSample* buf, unsigned int len)
{
buffer.put(ts,buf,len);
if(ts_less()(w_ts,ts+len))
w_ts = ts + len;
}
void AmPlayoutBuffer::buffer_get(unsigned int ts, ShortSample* buf, unsigned int len)
{
buffer.get(ts,buf,len);
if(ts_less()(r_ts,ts+len))
r_ts = ts + len;
}
//
// See: Y. J. Liang, N. Farber, and B. Girod. Adaptive playout scheduling
// and loss concealment for voice communication over IP networks. Submitted
// to IEEE Transactions on Multimedia, Feb. 2001.
// Online at:
// http://www-ise.stanford.edu/yiliang/publications/
// http://citeseer.ist.psu.edu/liang02adaptive.html
//
AmAdaptivePlayout::AmAdaptivePlayout(AmPLCBuffer *plcbuffer, unsigned int sample_rate)
: AmPlayoutBuffer(plcbuffer, sample_rate),
idx(0),
loss_rate(ORDER_STAT_LOSS_RATE),
wsola_off(WSOLA_START_OFF),
shr_threshold(SHR_THRESHOLD),
plc_cnt(0),
short_scaled(WSOLA_SCALED_WIN),
fec(sample_rate)
{
memset(n_stat,0,sizeof(int32_t)*ORDER_STAT_WIN_SIZE);
}
u_int32_t AmAdaptivePlayout::next_delay(u_int32_t ref_ts, u_int32_t ts)
{
int32_t n = (int32_t)(ref_ts - ts);
multiset<int32_t>::iterator it = o_stat.find(n_stat[idx]);
if(it != o_stat.end())
o_stat.erase(it);
n_stat[idx] = n;
o_stat.insert(n);
int32_t D_r=0,D_r1=0;
int r = int((double(o_stat.size()) + 1.0)*(1.0 - loss_rate));
if((r == 0) || (r >= (int)o_stat.size())){
StddevValue n_std;
for(int i=0; i<ORDER_STAT_WIN_SIZE; i++){
n_std.push(double(n_stat[i]));
}
if(r == 0){
D_r = (*o_stat.begin()) - (int32_t)(2.0*n_std.stddev());
D_r1 = (*o_stat.begin());
}
else {
D_r = (*o_stat.rbegin());
D_r1 = (*o_stat.rbegin()) + (int32_t)(2.0*n_std.stddev());
}
}
else {
int i=0;
for(it = o_stat.begin(); it != o_stat.end(); it++){
if(++i == r){
D_r = (*it);
++it;
D_r1 = (*it);
break;
}
}
}
int32_t D =
int32_t(D_r + double(D_r1 - D_r)
* ( (double(o_stat.size()) + 1.0)
*(1.0-loss_rate) - double(r)));
if(++idx >= ORDER_STAT_WIN_SIZE)
idx = 0;
return D;
}
void AmAdaptivePlayout::write_buffer(u_int32_t ref_ts, u_int32_t ts,
int16_t* buf, u_int32_t len)
{
// predict next delay
u_int32_t p_delay = next_delay(ref_ts,ts);
u_int32_t old_off = wsola_off;
ts += old_off;
if(short_scaled.mean() > 2.0){
if(shr_threshold < 3000)
shr_threshold += 10;
}
else if(short_scaled.mean() < 1.0){
if(shr_threshold > 100)
shr_threshold -= 2;
}
// need to scale?
if( ts_less()(wsola_off+EXP_THRESHOLD,p_delay) || // expand packet
ts_less()(p_delay+shr_threshold,wsola_off) ) { // shrink packet
wsola_off = p_delay;
}
else {
if(ts_less()(r_ts,ts+len)){
plc_cnt = 0;
buffer_put(ts,buf,len);
}
else {
// lost
}
// statistics
short_scaled.push(0.0);
return;
}
int32_t n_len = len + wsola_off - old_off;
if(n_len < 0)
n_len = 1;
float f = float(n_len) / float(len);
if(f > TSM_MAX_SCALE)
f = TSM_MAX_SCALE;
n_len = (int32_t)(float(len) * f);
if(ts_less()(ts+n_len,r_ts)){
// statistics
short_scaled.push(0.0);
return;
}
u_int32_t old_wts = w_ts;
buffer_put(ts,buf,len);
n_len = time_scale(ts,f,len);
wsola_off = old_off + n_len - len;
// if we have shrinked the voice, set back w_ts
// in order to have correct start point for possible
// PLC
if (n_len < (int32_t) len)
w_ts += n_len - len;
if(w_ts != old_wts)
plc_cnt = 0;
// statistics
short_scaled.push(100.0);
}
u_int32_t AmAdaptivePlayout::read(u_int32_t ts, int16_t* buf, u_int32_t len)
{
bool do_plc=false;
if(ts_less()(w_ts,ts+len) && (plc_cnt < 6)){
if(!plc_cnt){
int nlen = time_scale(w_ts-len,2.0, len);
wsola_off += nlen-len;
}
else {
do_plc = true;
}
plc_cnt++;
}
if(do_plc){
short plc_buf[FRAMESZ];
for(unsigned int i=0; i<len/FRAMESZ; i++){
fec.dofe(plc_buf);
buffer_put(w_ts,plc_buf,FRAMESZ);
}
buffer_get(ts,buf,len);
}
else {
buffer_get(ts,buf,len);
for(unsigned int i=0; i<len/FRAMESZ; i++)
fec.addtohistory(buf + i*FRAMESZ);
}
return len;
}
void AmAdaptivePlayout::direct_write_buffer(unsigned int ts, ShortSample* buf, unsigned int len)
{
buffer_put(ts+wsola_off,buf,len);
}
/**
* find best cross correlation of a TEMPLATE_SEG samples
* long frame
* * starting between sr_beg ... sr_end
* * to TEMPLATE_SEG samples frame starting from ts
*
*/
short* find_best_corr(short *ts, short *sr_beg, short* sr_end, unsigned int sample_rate)
{
// find best correlation
float corr=0.f,best_corr=0.f;
short *best_sr=ts;
short *sr;
for(sr = sr_beg; sr != sr_end; sr++){
corr=0.f;
for(unsigned int i=0; i<TEMPLATE_SEG; i++)
corr += float(sr[i]) * float(ts[i]);
if((best_sr == 0) || (corr > best_corr)){
best_corr = corr;
best_sr = sr;
}
}
return best_sr;
}
u_int32_t AmAdaptivePlayout::time_scale(u_int32_t ts, float factor,
u_int32_t packet_len)
{
// current position in strech buffer
short *tmpl = p_buf + packet_len;
// begin and end of strech buffer
short *p_buf_beg = p_buf;
short *p_buf_end;
// initially size is packet_len
unsigned int s = packet_len;
// we start from beginning of frame
unsigned int cur_ts = ts;
// safety
if (packet_len > MAX_PACKET_SAMPLES)
return s;
// not possible to stretch packets shorter than 10ms
if (packet_len < TEMPLATE_SEG)
return s;
if (fabs(factor - 1.0) <= SCALE_FACTOR_START) {
#ifdef DEBUG_PLAYOUTBUF
DBG("not scaling - too little f difference \n");
#endif
return s;
}
// boundaries of scaling
if(factor > TSM_MAX_SCALE)
factor = TSM_MAX_SCALE;
else if(factor < TSM_MIN_SCALE)
factor = TSM_MIN_SCALE;
short *srch_beg, *srch_end, *srch;
while(true){
// get previous packet_len frame + scaled frame
// (with size s) into p_buf
buffer_get(ts - packet_len, p_buf_beg, s + packet_len);
p_buf_end = p_buf_beg + s + packet_len;
// determine search region for template seg
// as srch_beg ... srch_end
if (factor > 1.0){
// expansion
srch_beg = tmpl - (int)((float)TEMPLATE_SEG * (factor - 1.0)) - SEARCH_REGION/2;
srch_end = srch_beg + SEARCH_REGION;
if(srch_beg < p_buf_beg)
srch_beg = p_buf_beg;
if(srch_end + DELTA >= tmpl)
srch_end = tmpl - DELTA;
}
else {
// compression
srch_end = tmpl + (int)((float)TEMPLATE_SEG * (1.0 - factor)) + SEARCH_REGION/2;
srch_beg = srch_end - SEARCH_REGION;
if(srch_end + TEMPLATE_SEG > p_buf_end)
srch_end = p_buf_end - TEMPLATE_SEG;
if(srch_beg - DELTA < tmpl)
srch_beg = tmpl + DELTA;
}
// stop if search region size < 0
if (srch_beg >= srch_end)
break;
// find best correlation to tmpl in srch_beg..srch_end
srch = find_best_corr(tmpl,srch_beg,srch_end,sample_rate);
// merge original segment (starting from tmpl) and
// best correlation (starting from srch) into merge_buf
float f = 0.5,v = 0.5;
for(unsigned int k=0; k<TEMPLATE_SEG; k++){
f = 0.5 - 0.5 * cos( PI*float(k) / float(TEMPLATE_SEG) );
v = (float)srch[k] * f + (float)tmpl[k] * (1.0 - f);
if(v > 32767.)
v = 32767.;
else if(v < -32768.)
v = -32768.;
merge_buf[k] = (short)v;
}
// put merged segment into buffer
buffer_put( cur_ts, merge_buf, TEMPLATE_SEG);
if (p_buf_end - srch - TEMPLATE_SEG < 0) {
ERROR("audio after merged segment spills over\n");
break;
}
// add after merged segment audio from after srch
buffer_put( cur_ts + TEMPLATE_SEG, srch + TEMPLATE_SEG,
p_buf_end - srch - TEMPLATE_SEG );
// size s has changed
s += tmpl - srch;
// go to next segment
cur_ts += TEMPLATE_SEG/2;
tmpl += TEMPLATE_SEG/2;
// calculate current factor
float act_fact = s / (float)packet_len;
#ifdef DEBUG_PLAYOUTBUF
DBG("at ts %u: new size = %u, ratio = %f, requested = %f (wsola_off = %ld)\n",
ts, s, act_fact, factor, (long)wsola_off);
#endif
// break condition: coming to the end of the frame (with safety margin)
if((unsigned int)(p_buf_end - tmpl) < TEMPLATE_SEG + DELTA)
break;
// streched enough?
if((factor > 1.0) && (act_fact >= factor))
break;
else if((factor < 1.0) && (act_fact <= factor))
break;
// streched over maximum already?
else if(act_fact >= TSM_MAX_SCALE || f <= TSM_MIN_SCALE)
break;
}
return s;
}
/*****************************************************************
*
* AmJbPlayout class methods
*
*****************************************************************/
AmJbPlayout::AmJbPlayout(AmPLCBuffer *plcbuffer, unsigned int sample_rate)
: AmPlayoutBuffer(plcbuffer, sample_rate)
{
}
u_int32_t AmJbPlayout::read(u_int32_t ts, int16_t* buf, u_int32_t len)
{
prepare_buffer(ts, len);
buffer_get(ts, buf, len);
return len;
}
void AmJbPlayout::direct_write_buffer(unsigned int ts, ShortSample* buf, unsigned int len)
{
buffer_put(ts, buf, len);
}
void AmJbPlayout::prepare_buffer(unsigned int audio_buffer_ts, unsigned int ms)
{
ShortSample buf[AUDIO_BUFFER_SIZE * 10];
unsigned int ts;
unsigned int nb_samples;
/**
* Get all RTP packets that correspond to the required interval,
* decode them and put into playout buffer.
*/
while (m_jb.get(audio_buffer_ts, ms, buf, &nb_samples, &ts))
{
direct_write_buffer(ts, buf, nb_samples);
m_plcbuffer->add_to_history(buf, PCM16_S2B(nb_samples));
/* Conceal the gap between previous and current RTP packets */
if (last_ts_i && ts_less()(m_last_rtp_endts, ts))
{
int concealed_size = m_plcbuffer->conceal_loss(ts - m_last_rtp_endts, (unsigned char *)buf);
if (concealed_size > 0)
direct_write_buffer(m_last_rtp_endts, buf, PCM16_B2S(concealed_size));
}
m_last_rtp_endts = ts + nb_samples;
last_ts_i = true;
}
if (!last_ts_i) {
return;
}
if (ts_less()(m_last_rtp_endts, audio_buffer_ts + ms))
{
/* Last packets have been lost. Conceal them */
int concealed_size = m_plcbuffer->conceal_loss(audio_buffer_ts + ms - m_last_rtp_endts, (unsigned char *)buf);
if (concealed_size > 0)
direct_write_buffer(m_last_rtp_endts, buf, PCM16_B2S(concealed_size));
m_last_rtp_endts = audio_buffer_ts + ms;
}
}
void AmJbPlayout::write(u_int32_t ref_ts, u_int32_t rtp_ts, int16_t* buf, u_int32_t len, bool begin_talk)
{
m_jb.put(buf, len, rtp_ts, begin_talk);
}