#1039 Higher quality audio

Mumble (514)

We are trying to use Mumble for voice lessons over VOIP.

For that, some sort of lossless codec would work very well. The issue is that opus even at max quality (96kbit) sounds very different than the direct output from the computer.
Maybe L16/48KHz/stereo or something along these lines?




  • Nicos Gollan
    Nicos Gollan

    Even if it was possible to integrate it with the current restrictions, the format you're proposing would mean 1500Kib/s and require around 190 packets per second, for each voice stream. Together with the multiplexing gain, that would put a very high load on the server and the clients.

  • Oh, I agree - but we are talking about 3-4 people listening, all on a LAN, or 2 people total, on a WAN.

    As another possibility, can we bump up the opus encoding bandwidth to way more than the 64kbit the VBR is selecting now? Maybe 192 or so. After all, if the coded is as good as vorbis, then at high bandwidth settings it should be pretty indistinguishable from raw PCM...

  • Stefan H.
    Stefan H.

    Lossless audio isn't an option for Mumble. We support Opus bitrates up to 96kbit/s which is near transparent in most situations. That's 96kbit/s mono, see the Paper Lights Opus sample http://opus-codec.org/examples/ for how far 96kbit/s can get you with stereo (need a recent firefox). We don't use VBR encoding though which degrades quality a bit but can't be helped as we need bandwidth to stay limited for our application.

    Most of the differences you'll hear between Mumble output and the original will be due to the aggressive noise filtering and agc Mumble performs. As we optimize more for speech and intelligibility rather than perfect reproduction this cannot be helped.

  • Nicos Gollan
    Nicos Gollan

    By the way, AGC and filtering can be disabled in the advanced audio input settings.

    IIRC it would be possible to use customised clients to send stereo audio with the Opus codec without breaking anything, as long as things stay inside the current bandwidth limits. That would of course also require implementing a non-mono audio input.

    For voice (I assume singing?), mono should be completely sufficient though, and with filtering disabled, the codec should do perfectly fine at the current rates.

  • Kissaki

    To come back to this, as there has been no activity for some time:

    From the initial request, I don’t quite get what “very different than the direct output from the computer” refers to.
    Directly outputting the mic input as audio output? So, I guess Mumbles filtering / improvements for voice (as mentioned before) are the issue here!?

    Did you try disabling the filtering as ngollan suggested?
    Looking at the audio input settings, “Noise Suppression” is the thing you want to turn down.

    Does this improve, or even fix your issue?

    As you mentioned and talked about opus specifically, I guess you are using snapshots?
    You made sure opus is used? By default Mumble will automatically fall back to a compatible codec if a client <= 1.2.3 connects after all.

  • Kissaki

    • status: pending --> wont-fix