audio artifacts sleuthing

Started by jbrower 5 years ago2 replieslatest reply 5 years ago118 views

I've uploaded a wav file ( that was produced by extracting payloads from an EVS (codec) pcap sent by our customer, and running the extracted file through the 3GPP reference decoder.  Similar (earlier) EVS pcaps from this customer sound fine, but those contained standard 3GPP reference content, not anything done "onsite".  In the new one, the voices of the guys talking sound muffled and fuzzy.  A spectrograph analysis doesn't show anything obvious like a repeating pattern (frame issue), excessive wideband noise, etc.

It sounds to me like the recording was in a room, using a PC type microphone, so maybe a softphone was used.  At least to me it doesn't sound like a pcap captured between two EVS enabled mobile phones.

Of course it could be an EVS payload (bitstream) content issue also, but in the EVS related work we've done the last couple of years I didn't yet hear an issue like this.

If anyone can take a listen and give any comment comments that come to mind, I would appreciate.  Thanks.


[ - ]
Reply by dgshaw6September 4, 2017

Which segments do you consider damaged?

The earliest utterances are definitely lacking in higher frequency components (muffled).

The four counting sequences in the middle are damaged by wind noise from the person being to close to and breathing into the mic.  Evidence is the very heavy low frequency content for the "four" and "five" which cause significant breath into the mic which damages the pitch content compared to adjacent segments.

It is also possible that the inputs to the encoder were saturated even though the output from the decoder is not.  This is evident, because the largest samples are well over 1/2 full scale, and the spectral content is significant at the frequencies above 4 kHz.

The last sequence, where the two are discussing "noise from A & B", are recorder some distance from the mic, so are a little muffled, but actually pretty resonable.

Hope this helps.


[ - ]
Reply by jbrowerSeptember 5, 2017

David - yeah it seems you're pretty good at artifact analysis.  After a lot of prodding we got more info from our customer and they did have issues with their test call setup.  In some cases they had two phones within close proximity of each other inside an electromagnetically shielded "chamber" of some type; in those cases when they put one phone on mute and repeated the tests audio quality cleared up considerably.  In other cases they were just setting the phone on the table and talking, without controlling their distance to the mic.

So I don't think they had wind noise, but I do think they could have got too close or too far wrt the mic.

My big worry was an EVS payload problem, for instance one phone's encoder not fully compliant with the EVS spec.  Most of the EVS enabled phones out there are using Qualcomm's 835 DSP core for media processing, which requires painstakingly optimized fixed point code, and the EVS code base is 3 separate libraries of over 300 total C files, so it's not out of the realm of possibility.  But I think for now this is ruled out.

Thanks very much for your time spent and your reply.