Noisy audio with upsampling in the conference bridge

[bennylp]

There are two problems here:

1. Problem with upsampling that introduces noise, and
2. Problem with resampling/upsampling, specific when it's used in the conference bridge and only when two or more sources are mixed together.

For the first problem:

The following file was upsampled with pjsua, from the original WAV file of 22050 Hz to 44100 Hz. You can hear clearly about the noise:

​http://www.pjsip.org/tmp/upsampling-noise.wav

For the second problem:

The following file was recorded by mixing two 16KHz sources and upsampled to 44.1KHz:

​http://www.pjsip.org/tmp/conference_noise.wav

[bennylp]

I identify so far the problem is related with the combination of clock_rate and ptime settings in the conference bridge. The resample algorithm is fine, because audio is playing fine with the resampleplay sample program.

So say for example, when clock_rate is 22050 and ptime is 10 msec (the default in pjsua_media.c), this gives samples_per_frame of 22050 * 10 / 1000 = 220.5, which is not a whole number! Setting the ptime to 20ms solved this for this particular clock_rate, but of course we'll still have this problem on 11025 clock rate.

Dunno yet what is the best solution for this, I guess temporarily we can use 20ms as the default ptime, and forbid users from using 11025 clock rate.

[bennylp]

In r906:
Changed PTIME in pjsua_media.c to 20ms (from 10ms).

[bennylp]

There is another problem, specific to when upsampling is done in conference bridge.

The conference uses the signal level of each sources as a parameter to the mixing algorithm. When the signal level changes significantly in the subsequent frame, this causes the subsequent frame to be misaligned with previous frame, and this would produce a noise.

This has been fixed in r931.

[bennylp]

(Forgot the credit)

Thanks Norman Franke of myasd.com for reporting, analyzing, and suggesting potential fixes for this.