Suggestion for AI denoising process

SpectraLayers
1

Hello, we were playing around with this old victrola recording:

The Unmix Song in Spectralayera 12 was able to separate the voice very well:

It was not able to separate the instruments from the victrola noise which I’m sure should be possible with further training of the Unmix AI model. This is just a suggestion for further SpectraLayer products.

Thank you!

2

Thx for the link.
Il Fantastico!

  1. Recording with Audacity-Loopback Device (…Steinberg should implement the loopback in all their programs…!!! This is no rocket science!)
    FFT-size 2048, Blackman-Harris
    48 kHz, 24 Bit
    image

  2. DeNoise

    image
    image473×244 11.5 KB

used noise gap

image
image270×763 54.2 KB

  1. DeClick
    image
    image384×216 11.7 KB

→ Sounds fine.

1 Like
3

thanks for sharing, yet, care to elaborate on the actual processes rendered? Just unmix song with voice checked? Or other preprocessing? Unmixnoisy speech and or DeNoise…DeClick?

so much more could be done (not that I wanna hear that song anymore :stuck_out_tongue_winking_eye:

again, certain unmixing at this current time is a starting point, albeit a very, very good starting point for further editing…and still, meat puppet manual editing is gonna arrive at an impressive finishing point which will require having to learn to identify what the spectrogram is displaying

4

Thank you for sharing @Sunnyman.
@ctreitzell there is no method, that is why I thought it was interesting. I just processed one time using Unmix Song, everything selected and you end up with most tracks silent, two are the voice and instrument/noise.