Conversation

Notices

Embed this notice
iced depresso (icedquinn@blob.cat)'s status on Friday, 11-Apr-2025 04:25:56 JST iced depresso

wonder how noise cancel shit actually works

do they just have a lab with a dummy and they run some big ML model to compute how to cancel all the sound

In conversation about 3 months ago from blob.cat permalink
- Embed this notice
  casey is remote (realcaseyrollins@noauthority.social)'s status on Friday, 11-Apr-2025 05:19:08 JST casey is remote
  in reply to
  
  @icedquinn Nowadays it's #AI
  In the olden days the only way to do noise cancelling as we view it today would be to have an exact sample of what you want to remove and invert it as inversion of audio waves creates silence. As tech got better, #AI could remove sounds similar to the sample you've provided, and nowadays you can use an #ML trained on a whole bunch of media that can remove stuff like vocals or music as needed
  I use this on occasion: https://vocalremover.org/
  In conversation about 3 months ago permalink
  Attachments
  1. Domain not in remote thumbnail source whitelist: vocalremover.org
    
    Vocal Remover and Isolation
    
    Separate voice from music out of a song free with powerful AI algorithms
- Embed this notice
  iced depresso (icedquinn@blob.cat)'s status on Friday, 11-Apr-2025 05:19:08 JST iced depresso
  in reply to
  - casey is remote
  @realcaseyrollins for microphones yeah. they asked people to go around with different mics and record ambient noise, then randomly overlaid it on studio recordings, that's how RNNoise and such was made.
  
  DeepFilterNet is stronger but a looot slower. I used that on my commercial audiobook productions.
  
  but i mean active hearing protection. it's a little harder since you have to calibrate the sound coming out of your drivers, which means like, you have to build a little silicone head or something to put good reference microphones on and then like
  
  yeah
  
  In conversation about 3 months ago permalink
- Embed this notice
  iced depresso (icedquinn@blob.cat)'s status on Friday, 11-Apr-2025 05:30:45 JST iced depresso
  in reply to
  - Password Man
  @mac_ack yea but you probably have to do slightly more than that idk
  
  In conversation about 3 months ago permalink
- Embed this notice
  Password Man (mac_ack@hidamari.apartments)'s status on Friday, 11-Apr-2025 05:30:46 JST Password Man
  in reply to
  
  @icedquinn they record ambient noise and play it back inverted
  
  In conversation about 3 months ago permalink
- Embed this notice
  iced depresso (icedquinn@blob.cat)'s status on Friday, 11-Apr-2025 06:02:14 JST iced depresso
  in reply to
  - Miander
  @mian i think i would just do it the way i was thinking where you make a reference lab box and just let the computer evolve or crunch something
  
  In conversation about 3 months ago permalink
- Embed this notice
  Miander (mian@mstdn.social)'s status on Friday, 11-Apr-2025 06:02:15 JST Miander
  in reply to
  
  @icedquinn
  the clever part is that the cost function is just the energy of the sound inside the headphones and the input is the ambient noise. Since the filter doesn't know the recording it can't cancel it.
  
  In conversation about 3 months ago permalink
- Embed this notice
  Miander (mian@mstdn.social)'s status on Friday, 11-Apr-2025 06:02:16 JST Miander
  in reply to
  
  @icedquinn
  IIRC they use a simple adaptive filter like
  https://en.wikipedia.org/wiki/Least_mean_squares_filter
  basically just one neuron
  In conversation about 3 months ago permalink
  Attachments
  1. No result found on File_thumbnail lookup.
    
    Least mean squares filter
    
    Least mean squares (LMS) algorithms are a class of adaptive filter used to mimic a desired filter by finding the filter coefficients that relate to producing the least mean square of the error signal (difference between the desired and the actual signal). It is a stochastic gradient descent method in that the filter is only adapted based on the error at the current time. It was invented in 1960 by Stanford University professor Bernard Widrow and his first Ph.D. student, Ted Hoff, based on their research in single-layer neural networks (ADALINE). Specifically, they used gradient descent to train ADALINE to recognize patterns, and called the algorithm "delta rule". They then applied the rule to filters, resulting in the LMS algorithm. Problem formulation The picture shows the various parts of the filter. x {\displaystyle x} ...
- Embed this notice
  iced depresso (icedquinn@blob.cat)'s status on Friday, 11-Apr-2025 06:02:57 JST iced depresso
  in reply to
  - Miander
  @mian to an extent this also works for materials but eh. i don't know vibration specialists and shit like ploopy does.
  
  In conversation about 3 months ago permalink

Public

Conversation

Notices

Feeds