Learn how easy it is to sync an existing GitHub or Google Code repo to a SourceForge project! See Demo


Voice Morph

  • Is it possible to configure murmur (server-side) so that all voices are morphed (without mumble control to turn it off) instead of having to install a morph-app (like MorphVox) on the client. We're trying to setup a secure chat client for WoW but we don't want anyone to be able to ID the users since we have a number of kids in the guild.


    • M^3

      I would think this would be a per-channel setting, not a global setting, but whatever works for the developer is fine w/ me.

    • This would require server-side processing of the voice audio, which would mean transcoding, which means reduced quality and increased latency. So it's not going to happen.

      • > This would require server-side processing of the voice audio,

        Not necessarily - the client side could do the voice morphing (maybe add it as an additional option). The server could enforce it by ACL's.

        But you are right - even on the client side the processing would add to the latency.
        The quality loss won't be a problem, because the whole point of voice morphing is, that the quality is "different" than before.

        Coming up with some good algorithms that do the morphing might be a tricky one.

        • Doing it on the client wouldn't ensure any kind of security. It would not be "secure", it would merely make it inconvenient to bypass it. Anyone who wish to bypass it are free to do so by recompiling the client. So in the end, you end up with something that doesn't really enforce anything but a false sense of security.

          Besides, anyone with basic psychology skills will be able to discern the "true age" of the speaker based on speech patterns and responses to questions and phrases. Start talking about one-hit radio wonders that faded away with time. If the answer is "huh? who?" you just found the upper limit of their age.

          Similarily, anyone who has worked a bit with voice codecs will be able to do noise analysis on the "morphed" audio to discern the true frequency distribution of the original voice. This will not be enough to reproduce the original voice, but will be more than enough to say that "no 12 year old girl has a voice that deep".

          So voice morphing is "cool", and it's enough to distort your voice so no exact matching is possible, but please don't think it will mask out all information about who you are.

        • Doing the work on the client side and having some ACL setting would be fine. There are plenty of morph packages out there that could be used IF the client had a plugin structure that would let us specify required functions from the server-side. The interesting thing about this is that no commercial app appears to have this functionality either. Any chance the client/server communications supports enough bi-directional communication to setup a security/ACL feature like this?


    • M^3

      you could always tap into the speech recognition engine in windows xp (or some office releases if installed) and turn it into text, then have the test output as speech at the receiver. Not a good idea if playing something noisy or something that requires a lot of CPU that would be eaten up by the speech recognition engine, the latency would be horrible as well, and many other short-comings as well, but you could do it. Oh, then while you are at it have a text filter for age related references. OK, so i am rambling pointlessly now. So Sue me.