8 Replies Latest reply: Aug 30, 2014 5:33 AM by ryclark RSS

    Sound removal problem with speech


      I have started to use a T-bone 800 microphone. I have Audition CS6 on an iMac and to get the microphone into the computer directly, I use a multimix 8 USB unit.


      The problem is that the Multimix 8 adds a slight but annoying whistle to the recording.


      Using spectral view, I can identify the whistling noise (it manifests as distinct bands on distinct frequency ranges). And I can capture a noise profile and apply it to the whole file. Which works well. EXCEPT ...


      I use Audition for recording people talking. I recorded a guy reading stuff for 20 minutes and the problem I have is that although the noise reduction works perfectly when he isn't talking, you can still hear the whistle (reduced in volume but still audible) when he is talking. I can even see the ghost of the whistle on spectral view, even though this vanishes on the audio where he stops talking for more than a quarter of a second.


      I have tried to tackle the problem differently, with a graphic equaliser and noise reduction and had some success. But I'm a complete amateur when it comes to things like noise reduction!


      So can I ask why the noise reduction fails to subtract the noise during speech? And how would an expert set about removing this annoying whistle (and very off topic but if anyone knows why the Multimix 8 adds a whistle in the first place, I'd be grateful to find out! I did once record by channeling the t-bone through my Marantz 660 recorder and that works but is a clumsy solution).


      Someone airily told me to use the parametric equaliser but I'm out of my depth!

        • 1. Re: Sound removal problem with speech
          SteveG(AudioMasters) Community Member

          We'd need a short sample of the noise with the speech before we could tell you the best way to eliminate it. Being an audio forum, of course you can''t post audio on it (thanks, Adobe!) but a Dropbox link would work fine.

          • 2. Re: Sound removal problem with speech
            Pentatonia Community Member

            Wow, hadn't expected such instant and helpful responses! I'm currently uploading. Presumably I can paste the dropbox link here

            • 3. Re: Sound removal problem with speech
              Pentatonia Community Member

              Hopefully this will work!!


              Dropbox - soundproblem.wav

              • 4. Re: Sound removal problem with speech
                SteveG(AudioMasters) Community Member

                Download works fine. The noise is interesting, and isn't going to be removed by normal EQ at all. Something, and I don't know what, is producing a 1kHz tone with harmonics every 1kHz above it at a similar level. And there are loads of them - I've counted 22, right up to the Nyquist frequency. It's not just the Multimix 8 that does this - there's a guy on Youtube complaining about exactly the same thing with a Multimix 4, so I'd say that it's very likely to be a design or construction flaw, almost certainly in the area concerned with connecting to your computer.


                Probably, to get rid of this from a recording, the best way to proceed is to use more than one pass of the notch filter, with it set to 1kHz increments until you run out of channels. Then start again with the next set of frequencies going up - you'll probably need 3+ passes to do this. The exact settings you'll need to experiment with, but these are sharp spikes, so probably very narrow settings. If you open Audition's frequency analysis window and just analyse the quiet bit at the end of your file, you'll see them clearly. You will have to be a bit careful though, because they are present throughout the band you want to retain, which always causes rather more problems for the treatment. What you might well be able to do when you've found a series of settings that work is to save each one, and then deploy then in a sequence as a favourite. Because of the very peculiar nature of this, I really don't think anything else is going to help.


                NR won't work very well, because that's based on statistical noise reduction rather than tonal treatments. Software like iZotope's RX3 can deal better with this sort of problem, but it's expensive. You'd be better off putting the money towards an improved interface, I'd say...

                • 5. Re: Sound removal problem with speech
                  ryclark Community Member

                  I successfully used the DeHummer effect with settings as shown in the picture. A bit more tweaking of attenuation levels might be better with less. It did still leave a few high harmonics above about 10K but they were fairly down in the noise at those frequencies.


                  ScreenHunter_31 Aug. 29 23.53.jpg

                  • 6. Re: Sound removal problem with speech
                    Pentatonia Community Member

                    WOW, I feel humbled to have received so much help! Steve has put his finger on the cause, in that it does seem to be a problem with the device and I'm pondering where to go with that. But Ryclark's use of the dehummer is astonishingly good - I've only just done it and on first hearing, it is absolutely perfect (and the file I applied it to is much bigger than the one I sent of course)! I guess I'll need to knuckle down and find out how to use the dehummer from scratch, as it were (if you have time, perhaps you could explain how you arrived at the dehumming settings, because I imagine I'm not the only idiot who has never used this function!). But a huge thanks to both of you!

                    • 7. Re: Sound removal problem with speech
                      SteveG(AudioMasters) Community Member

                      Just in case it's not clear, DeHummer = Posh Notch Filter!


                      The basic approach is to look at the strength of the harmonics using the frequency analyser (Window>Frequency Analysis) as a start. Set the base frequency to the lowest spike (zoom in on the analyser window to set this accurately). For thin spikes, set the Q value quite high - although you have to bear in mind that this can sometimes cause audible anomalies, so be careful with that. Set the gain to approximately the height of the spikes above the noise floor, and the slope so that it looks like this inverse of the slope of the spikes you've got. You may have to fine-tune everything except the base frequency.


                      What you are looking for in the Frequency Analysis window is for nothing to go below the noise floor. That would definitely indicate that you'd overdone it! Ultimately though, your ears should be the final arbiter of how far you go with this.

                      • 8. Re: Sound removal problem with speech
                        ryclark Community Member

                        Although, as it's name suggests, the DeHummer effect is meant for getting rid of low frequency hum and it's harmonics it is also a very useful notch filter.


                        You can do some fine tuning by ticking the Output Hum Only box to just listen to the hum and making adjustments to the fundamental frequency.