I have the following audio file : https://vocaroo.com/1nF67OiFA9hm
I'd like to know what kind of distortions are present so that I can reproduce those on other "clean" speech.
Thanks
2 Answers
It looks like two unbalanced signals have been added together by just joining the wires together. The + sides added have hit the power rail and become flattened. The - sides added have made an "extra negative" part of the signal and are twice as extreme. The way both sides accelerate towards a sharp point indicates a pre-amp with no feedback control running at the end of it's linearity (too hot). It causes a runaway increase / sharpens the curve near the top. Most sine waves should be nearly flat at the top not spikey.
It sounds like you want to develop a convolution filter. However, to do so, you need both the input and the output, or in some cases, and large patch of silence from the same environment, eg this file but with extra space included called the noise floor sample.
Anyhow looking at the file in Audacity it appears quite asymetric:
Most sounds appear like a mirror or reflection about zero, yours goes from +0.3 to -0.8. On my system there is a click when playback finishes but silence should not click when stopped. That indicates a DC offset.
If I extract a piece of silence and normalize without adjust DC
I see that most of the signal stays in the + side. That is bad.
If I choose to adjust the DC offset around zero (normally what you want) it starts to look a bit more normal:
Choosing Plot Spectrum I can see a sharp roll off on all the trebles above 7khz. To recreate this add a shelving filter / low-pass filter at 7581 hz.
Did you know that audio electronic circuits can only process positive voltages? So if your line level goes from +1 volt to -1 volt, that's 2 volts peak to peak. Say a 1 volt bias is added, the signal is now from+2 to 0 volts, but still 2 volts peak to peak: it can now be processed using a single sided amp. That's how unbalance line level signal paths work as far as I know. It saves a fair bit of money when building component HiFi systems.
But this is why you can't mix audio by connecting two sets of RCA cables together. /tmp/Spectacle.yjpzUG/Screenshot_20230112_024624.jpg
One idea would be using a vocoder. I am using the Charlie Brown's teacher trombone sound from Snoopy as a carrier.
You can do this via panning and then select the option L R Encoding in Vocodex in FL Studio.
