Hello SignalProcessing!

Given the important frequencies of a human voice (360Hz to 5kHz) to communicate and the frequencies above and below that make the voice sound way better.

Imagine we are recording a phone call... the provider normally transmitts only the frequencies needed for moderate communication. I am interested in reconstructing the information for the lost frequencies above and below by training deep neuronal nets. To improve the training progress it would be helpful to figure out if there is sort of a correlation.

Any reference about that topic would be great! Thank you.

  • $\begingroup$ "Is there a known correlation between human voice (360Hz to 5kHz) and the frequencies above/below that?" yep, speech models have done that training for you, decades ago :) "To improve the training progress it would be helpful to figure out if there is sort of a correlation." LOL. Can we agree that without correlation, your neural network couldn't do sh.. err, anything about the signal? $\endgroup$ – Marcus Müller Mar 26 '18 at 14:04
  • $\begingroup$ Hey Marcus, you probably know way more than i do and i am really interested to get into this topic! I really would love to get some advice about papers regarding to speech models? I am new to that, but also happy i made you sort of laugh out loud! $\endgroup$ – Chrizzldi Mar 27 '18 at 21:09

There is a correlation and it can be learned with machine learning methods. A simple google search can give you many papers, recent research

Waveform Modeling and Generation Using Hierarchical Recurrent Neural Networks for Speech Bandwidth Extension Zhen-Hua Ling, Yang Ai, Yu Gu, Li-Rong Dai

older paper:

Techniques for the Regeneration of Wideband Speech from Narrowband Speech Jason A. Fuemmeler, Russell C. Hardie, William R. Gardner


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.