pjsip的语音检测

转自http://blog.sina.com.cn/s/blog_513f4e8401011hf9.html

在pjsip中包含了语音静音检测的实现。
静音检测开关宏

/** * Specify how long (in miliseconds) the stream should suspend the * silence detector/voice activity detector (VAD) during the initial * period of the session. This feature is useful to open bindings in * all NAT routers between local and remote endpoint since most NATs * do not allow incoming packet to get in before local endpoint sends * outgoing packets. * * Specify zero to disable this feature. * * Default: 600 msec (which gives good probability that some RTP  *                    packets will reach the destination, but without *                    filling up the jitter buffer on the remote end). */#ifndef PJMEDIA_STREAM_VAD_SUSPEND_MSEC#   define PJMEDIA_STREAM_VAD_SUSPEND_MSEC600#endif

方法就是对每帧的采样点进行累加，得到的结果和门限进行比较。

语音检测分为固定模式和动态模式。固定模式就是累加值和给定的门限比较，动态模式的具体方法还不确定。
语音检测的调用关系：

g711_encode()->pjmedia_silence_det_detect（）g711_encode：is_silence = pjmedia_silence_det_detect(priv->vad,(const pj_int16_t*) input->buf,(input->size >> 1), NULL);if (is_silence &&(PJMEDIA_CODEC_MAX_SILENCE_PERIOD == -1 ||silence_period < PJMEDIA_CODEC_MAX_SILENCE_PERIOD*8000/1000)){output->type = PJMEDIA_FRAME_TYPE_NONE;output->buf = NULL;output->size = 0;output->timestamp = input->timestamp;return PJ_SUCCESS;} else {priv->last_tx = input->timestamp;}}........

这里把代码贴一下：

PJ_DEF(pj_bool_t) pjmedia_silence_det_apply( pjmedia_silence_det *sd,pj_uint32_t level){int avg_recent_level;if (gp_pjmedia_silence_det->mode == VAD_MODE_NONE)//无模式return PJ_FALSE;if (gp_pjmedia_silence_det->mode == VAD_MODE_FIXED)//静态模式return (level < gp_pjmedia_silence_det->threshold);)//动态模式gp_pjmedia_silence_det->sum_level += level;++gp_pjmedia_silence_det->sum_cnt;avg_recent_level = (gp_pjmedia_silence_det->sum_level / gp_pjmedia_silence_det->sum_cnt);if (level > gp_pjmedia_silence_det->threshold ||level >= PJMEDIA_SILENCE_DET_MAX_THRESHOLD){gp_pjmedia_silence_det->silence_timer = 0;gp_pjmedia_silence_det->voiced_timer += gp_pjmedia_silence_det->ptime;switch(gp_pjmedia_silence_det->state) {case STATE_VOICED:if (gp_pjmedia_silence_det->voiced_timer > gp_pjmedia_silence_det->recalc_on_voiced) {gp_pjmedia_silence_det->threshold = (avg_recent_level + gp_pjmedia_silence_det->threshold) >> 1;TRACE_((THIS_FILE,"Re-adjust threshold (in talk burst)""to %d", gp_pjmedia_silence_det->threshold));gp_pjmedia_silence_det->voiced_timer = 0;gp_pjmedia_silence_det->sum_level = avg_recent_level;gp_pjmedia_silence_det->sum_cnt = 1;}break;case STATE_SILENCE:TRACE_((THIS_FILE,"Starting talk burst (level=%d threshold=%d)",level, gp_pjmedia_silence_det->threshold));case STATE_START_SILENCE:gp_pjmedia_silence_det->state = STATE_VOICED;gp_pjmedia_silence_det->sum_level = level;gp_pjmedia_silence_det->sum_cnt = 1;break;default:pj_assert(0);break;}} else {gp_pjmedia_silence_det->voiced_timer = 0;gp_pjmedia_silence_det->silence_timer += gp_pjmedia_silence_det->ptime;switch(gp_pjmedia_silence_det->state) {case STATE_SILENCE:if (gp_pjmedia_silence_det->silence_timer >= gp_pjmedia_silence_det->recalc_on_silence) {gp_pjmedia_silence_det->threshold = avg_recent_level << 1;TRACE_((THIS_FILE,"Re-adjust threshold (in silence)""to %d", gp_pjmedia_silence_det->threshold));gp_pjmedia_silence_det->silence_timer = 0;gp_pjmedia_silence_det->sum_level = avg_recent_level;gp_pjmedia_silence_det->sum_cnt = 1;}break;case STATE_VOICED:gp_pjmedia_silence_det->state = STATE_START_SILENCE;gp_pjmedia_silence_det->sum_level = level;gp_pjmedia_silence_det->sum_cnt = 1;case STATE_START_SILENCE:if (gp_pjmedia_silence_det->silence_timer >= gp_pjmedia_silence_det->before_silence) {gp_pjmedia_silence_det->state = STATE_SILENCE;gp_pjmedia_silence_det->threshold = avg_recent_level << 1;TRACE_((THIS_FILE,"Starting silence (level=%d ""threshold=%d,g_silence_nbr:%d)", level, gp_pjmedia_silence_det->threshold,g_silence_nbr));gp_pjmedia_silence_det->sum_level = avg_recent_level;gp_pjmedia_silence_det->sum_cnt = 1;}break;default:pj_assert(0);break;}}return (sd->state == STATE_SILENCE);}