On 15.02.2018 10:31, Raman Shishniou wrote: > On 02/15/2018 10:20 AM, Georg Chini wrote: >> On 14.02.2018 22:51, Raman Shishniou wrote: >>> On 02/14/2018 07:20 PM, Georg Chini wrote: >>>> On 14.02.2018 15:25, Raman Shyshniou wrote: >>>>> This patch adds a underrun_protection argument to >>>>> control underrun protection algorithm. Disabling >>>>> protection will keep loopback latency regardless >>>>> of underruns. >>>> Again I do not understand the motivation of the patch. >>>> In what situations are you seeing so many underruns and >>>> still want to keep the original configured latency value? >>>> Audio will be very bad in that case. >>>> >>> All situations where where latency is more important than data integrity. >>> Voice over IP (telephony) for example, receiving audio data using network >>> by UDP/RTP. Any data loss leads to underruns in loopback module. >> This is not correct. It will only lead to underruns, if module-loopback runs >> out of data. So if you buffer enough data, missing packets in the voice >> stream would just appear to be small sample rate variations from the >> perspective of the loopback module. Because the loopback module >> does some adaptive re-sampling, these variations are no problem. >> >> Maybe this happens for you because module-pipe-source has no buffer >> at all and simply passes the values through whenever data arrives. >> With missing data packets, this can surely lead to underruns on the >> source side which are passed on to the loopback module. >> Perhaps you should implement some buffering in pipe-source? >> >> I don't see the point of your change. If you are seeing massive underruns, >> audio quality will be really bad and just sticking to the configured latency >> will not improve the situation. For me, the permanent occurrence of >> underruns shows that there is something wrong with your setup in the first >> place. The better idea is to correct the audio chain, so that no (or only very >> few) underruns happen. > I Agree. For live streaming or internet radio I can buffer more data up to > several seconds (or just use tcp). But not for telephony, where 10-20 > underruns per hour is acceptable, but latency more than 50-100ms is not. > Loopback module increases latency permanently. There is no way to decrease > it without unload and load again. How about a max_latency_msec argument in the sense that the logic will not be completely disabled, but module-loopback will not try to increase the latency above max_latency_msec (if specified) even if underruns occur?
