Forums

voice changer

Started by python November 15, 2006
is it possible to change someone's voice to sound like a particular
person's voice? e.g make my voice sound like George Bush(accent, pitch,
length of words and all). i'm thinking of doing this as a university
project but i am completely new to this field of dsp. are there algorithms
for doing this? please advice me.



python wrote:
> is it possible to change someone's voice to sound like a particular > person's voice? e.g make my voice sound like George Bush(accent, pitch, > length of words and all). i'm thinking of doing this as a university > project but i am completely new to this field of dsp. are there algorithms > for doing this? please advice me.
In principle, it shoule be possible. There are text-to-speech engines that have voice "personalities" built in. The resulting voice sounds like a person. I suppose you could build a George Bush personality or whatever. Then, you could have a good speech recognition engine that converts your speech to text and run it through the TTS. That said, think of the speech recognition engine and the TTS engine in terms of reciprocal hierarchical blocks and you could find an intermediate point to connect the two. Your speech can be characterized in terms of some parameters which are then passed through a Bush transform and played out as the modified speech. I know of no literature that deals with this problem though. C
>I suppose you could build a George Bush personality or >whatever. Then, you could have a good speech recognition engine that >converts your speech to text and run it through the TTS.
I would think that preceding this with a random text generator would produce a convincing result.
Jeff Caunter wrote:

>>I suppose you could build a George Bush personality or >>whatever. Then, you could have a good speech recognition engine that >>converts your speech to text and run it through the TTS. > > > I would think that preceding this with a random text generator would > produce a convincing result. >
It would be quite a challenge to emulate the characteristic slurred speech of the (reformed ?) alcohol and cocaine abuser. Paul
I must add do not forget to come up with some stupid sentences if you 
decided to chose Bush, since I assume that you are smarter than him.


"Paul Russell" <prussell@sonic.net> wrote in message 
news:4s0r62Ft468vU1@mid.individual.net...
> Jeff Caunter wrote: > >>>I suppose you could build a George Bush personality or >>>whatever. Then, you could have a good speech recognition engine that >>>converts your speech to text and run it through the TTS. >> >> >> I would think that preceding this with a random text generator would >> produce a convincing result. >> > > It would be quite a challenge to emulate the characteristic slurred speech > of the (reformed ?) alcohol and cocaine abuser. > > Paul
"Chetan Vinchhi" <vinchhi@gmail.com> wrote in message
news:1163601856.896990.291820@f16g2000cwb.googlegroups.com...
> > python wrote: > > is it possible to change someone's voice to sound like a particular > > person's voice? e.g make my voice sound like George Bush(accent, pitch, > > length of words and all). i'm thinking of doing this as a university > > project but i am completely new to this field of dsp. are there
algorithms
> > for doing this? please advice me. > > In principle, it shoule be possible. There are text-to-speech engines > that have voice "personalities" built in. The resulting voice sounds > like a person. I suppose you could build a George Bush personality or > whatever. Then, you could have a good speech recognition engine that > converts your speech to text and run it through the TTS. > > That said, think of the speech recognition engine and the TTS engine in > terms of reciprocal hierarchical blocks and you could find an > intermediate point to connect the two. Your speech can be characterized > in terms of some parameters which are then passed through a Bush > transform and played out as the modified speech. I know of no > literature that deals with this problem though. > > C >
To make a George Bush system just use a system that generates random amounts of crap with key words 'Iraq',WMDs etc. M. -- Posted via a free Usenet account from http://www.teranews.com
Major Misunderstanding wrote:
> "Chetan Vinchhi" <vinchhi@gmail.com> wrote in message > news:1163601856.896990.291820@f16g2000cwb.googlegroups.com... >> python wrote: >>> is it possible to change someone's voice to sound like a particular >>> person's voice? e.g make my voice sound like George Bush(accent, pitch, >>> length of words and all). i'm thinking of doing this as a university >>> project but i am completely new to this field of dsp. are there > algorithms >>> for doing this? please advice me. >> In principle, it shoule be possible. There are text-to-speech engines >> that have voice "personalities" built in. The resulting voice sounds >> like a person. I suppose you could build a George Bush personality or >> whatever. Then, you could have a good speech recognition engine that >> converts your speech to text and run it through the TTS. >> >> That said, think of the speech recognition engine and the TTS engine in >> terms of reciprocal hierarchical blocks and you could find an >> intermediate point to connect the two. Your speech can be characterized >> in terms of some parameters which are then passed through a Bush >> transform and played out as the modified speech. I know of no >> literature that deals with this problem though. >> >> C >> > To make a George Bush system just use a system that generates random amounts > of crap with key words 'Iraq',WMDs etc. > > M. > > >
And of course FREEDOM! To get back to the original question, there seems to be two parts to the solution. The first and easiest is to take your speech and change it so that it sounds as though it has come from Bush's vocal tract. For example, DSP can be used to equalise the effect of existing vocal tract resonances and to insert new resonances. If this processing is done well, the new voice signal might pass a voice-print test but it would not yet fool anyone into thinking it was coming from George W. himself. The second and hardest part is to identify and reproduce the words, accent and mannerisms that make W's speech so recognisable. This really needs a skilled mimic. Interestingly, people who can do this well are judged to sound very much like their subject even though their vocal tract may not be at all similar. I don't know how you would even start to tackle this mimicking process electronically, however my humble suggestion for doing it non-electronically is this: 1. Practice the accent, and when you have it down really well, start on the mannerisms. 2. Write out some text that is full of crap, as has been suggested, but take care to choose your words from the same limited vocabulary as George W. 2. Display the text on a teleprompter. If you don't have a teleprompter, print it out in a really large font. 3. Read the text out as though you are seeing it for the first time, raising your inflection as you approach the end of each line regardless of the sentence structure. 4. Look up from time to time to see if everyone is a puzzled as you. 5. Be sure to say 'freedom' a lot. Regards, John