3

Is there a way to create an utterance in synthetic speech which is devoid of prosody? Or if no such synthetic speech system exists, would there be a way to filter an existing utterance to remove the intonation contours?

3
  • Now that you mention it, I remember that even Eloqmini or Loquendo have some (even if weird) forms of intonation and pitch. They sound artificial, yes, but they are there nonetheless.
    – Alenanno
    Sep 27 '12 at 8:17
  • 1
    Do you just want speech that has a constant F0? Like literally monotone? If that's the case, you can just read any speech into Praat and create a manipulation object, and remove all but a single pitch point. Sep 27 '12 at 13:05
  • Thanks, @musicallinguist -- that's what I ended up doing, but it wound up with some weird blips and cracks, as if someone was speaking into a microphone. I think this should be easy to filter with other editing software, but let me know if you have any further thoughts on removing those.
    – Julie
    Sep 28 '12 at 0:37
3

I was going to continue the thread of my comment under the original question, but I figured I might as well turn it into an answer and not have to deal with the word limit!

There are so-called "rule-based" speech synthesis systems out there that can yield output with any kind of desired F0, but it sounds like what you are really looking for is a commercial text-to-speech program that takes text as input and allows you to control the pitch of the output. The important distinction here is that in order to use the former you need to know how to do speech synthesis from "scratch" and manipulate all of the relevant parameters, which I'm assuming is not the case for you! As for the latter, I'm not aware of any such systems that are commercially available at this time.

So, an alternative solution is to use the Praat Manipulation features. Here are a couple of tips to minimize distortion in the manipulated tokens:

  • You'll want to avoid manipulating the F0 of segments that contain voiced frication noise, like [z] and [v]. The problem with such segments is that they contain voicing with an audible pitch that you'd want to manipulate, but the manipulation doesn't work so well when there is frication noise mixed in and you get some weird distortions as a result. I'm not sure if you have control over the segmental makeup of the stimuli you are creating, but if you do it's best to construct utterances that don't contain such segments.
  • If possible, you should avoid manipulating speech that has extreme F0 excursions (i.e. a very wide pitch range) to begin with, as the more you "stretch" the pitch of speech, the less natural it will sound.
  • You'll want the input recordings (if they are indeed recordings) to be as clean as possible, with minimal external noise.

I hope this does the trick! If you have problems with unwanted distortions or noise, let me know some specifics (like what kind of noise and the context in which it is occurring) and I'll try to recommend ways of avoiding it or getting rid of it after-the-fact).

1
  • interesting to know!
    – user483
    Sep 29 '12 at 16:17

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.