How to get rid of this distortion of speech?
Here is a short example
One could do this manually, of course, but that would take days…
We have several hundreds of those files and they are long… this is just one of them.
wow it is pretty severe! luckly it is also pretty distinct and pretty homogeneous.
My absolute first step would be to start a batch process on all files with Unmix Noisy Speech + Unmix Components on each layer for good measure.
The Components Unmixing might not do much good, but I figure since the noise is so tonal in nature, it can be useful for recovering some mouth noises that might be mis-interpreted by UNS. So might as well do it at once.
Then I’d go on vacation and see what I get when I come back. =)
Same clip after Unmix Noisy Speech.
I tried Unmix Song, Vocals, quality “Balanced”:
Must be a definitions thing. I don’t hear distortion, I hear a straightline buzz interferring with the relatively lo-fi speech underneath.
The posted experiment test results I hear so far have a lot of artifacts, and the buzz still faintly there.
Hundreds of clips like that?
I subsequently did a DeBleed to remove voice remnants from the buzz. The vocal was slightly improved, but not such that I felt the need to post it.
The quality isn’t great, but the original is a mess!
Thx a lot for your comments, they help a lot.
The noise is electric interference, very regular.
Any idea, what might have caused that? Maybe an electrical engineer could shine some light on this? It sounds similiar to a triangle wave.
Almost certainly a faulty microphone, or microphone cable or connector. The interference is primarily 50Hz mains hum, with its many harmonics. You can get the same interference by plugging a guitar cable into an amp and touching the tip with your finger (60Hz in the US and some other countries, 50Hz in the EU/UK).
… I think the harmonics of 50Hz might be the result of the interference signal itself over-driving the mic input.
I think the results are at least intelligible, which could be enough for e.g. transcription, so there’s also a question of what the deliverable should be used for. “Unmix Noisy Speech” might be better for transcription, whereas “Unmix Song | Vocals” might be better for video post with ambience (notice the glasses clinking, and the cough in the middle). Should the result be in stereo or mono?
I’m sure improvements could be made by further processing with other tools to deal with room resonance etc., but I think the objective is to do it within SpectraLayers as part of a batch process.
At the start of the piece, in my example, those are not artifacts, it’s liquid being poured into a glass. You can hear the clink of the glass as the decanter or bottle touches the rim of the glass.
From my perspective, this is an example of the type of job I would not have taken on prior to SpectraLayers, as the prospect of success would have been dismal.
Thx a lot!!!
Restored in iZotope.
- De-Hum in Adaptive Mode (Sensitivity 7.5 / Bands 1020 / Filter Q 1000)
- Spectral Denoise at 6dB in Highest Quality mode and reduction curve at the low end
- De-click (Multi-band random click / Sensitivity 8.0 / Freq skew 10.0 / Click widening 0.0)
- dxRevive Pro with Natural Algorithm at 50%
Could be automated easily.
In SL13 using:
Unmix Noisy Speech>
Unmix Components (Noise Layer)>
-130dB level change (Tonal Layer)>
dxRevive Studio 3 algo at 20%.
If I could understand the language I’d possibly tweak things differently.
It’s German but even I as a native speaker am having a hard time to understand what is said.
Here what I did was;
1- Unmix components. On the tonal layer, register noise with denoise module and take it out %100 percent. then merge back.
2- Unmix noisy speech, on noise layer apply -6 db gain on mid highs and highs with a genle fade to around 2k hz.
3 - Voice dereverb % 70 reduction
4 - -6 db gain reduction in between 4k - 7k hz again with a fade
5 - dxRevive Pro Studio 3 algorithm, %60. (I don’t think you can do this without dxRevive. These are very dxRevive territory problems)
After this point I’d do manual cleanup but it’d take a lot of time if you go that way