I have built a text to speech system that is generating speech for a given text, however the generated audio has a lot of vibrations in it. Although the speech is clear, it doesn't really sounds natural and is a bit robotic.

Here is a sample of the generated audio: Audio

Please give some suggestion on how I can clean this audio and make it sound more natural.


You should investigate the overlap-add technique of merging your audio buffers. It sounds to me that the buffers are windowed but are possibly just played out in sequence, causing the 'vibration' you refer to. Using an overlap add technique you should be able to smooth this out. Also check out the code for 'paulstretch' as this uses a similar synthesis technique to stretch out audio over long timeframes.


It seems like your sample rate is very low. That is handy for keeping sizes low, but it ruins the output (unless you like robot speech)

Try higher quality samples and a much higher sample rate.

  • The sample rate is at 22khz. Is there a way I can modify this audio to make it more natural sounding? Jan 14 '19 at 23:47
  • Weird. It sounds much lower. I'd have guessed half that from your audio clip
    – Rory Alsop
    Jan 15 '19 at 13:27
  • The audio is computer generated, so that might be why it seems the sample rate is low. Feb 26 '19 at 12:51

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.