speech analysing

Is it possible to convert one persons voice to another persons voice..?? How..??

Answer 1

[Daniel Shub]

It is not possible to convert one persons voice to another persons voice. It is hard enough to even characterize what one persons voice is.

[Walter Roberson]

I do not know if I would go as far as saying that it is not possible. Difficult, certainly.

Suppose you had a speech recognition system that was able to recognize phonemes with relatively high accuracy, and you put that through a speech-to-marked-up-text system. And then suppose you took that marked-up-text and put it through a voice synthesis system based upon a different person. The result would be that the speech had been received in one person's voice, but would be output in a different person's voice.

Meanwhile, the bus-stop announcement system on our local buses drives me bonkers as I try to figure out why so many stops are mispronounced. Like "Navy" getting pronounced as "neh-vee" instead of "nAv-E" where A is long-A and E is long-E. Not much difference in the phonemes, mostly difference in the durations and clipping, but it is enough to make street names unrecognizable.

Even the main street of the city gets mispronounced by the automated system, half of it pronounced as in French and half pronounced as in English, and so jarring for both our main linguistic groups.

[Daniel Shub]

The problem is the voice synthesis. I don't know of any "good" voice synthesis systems (although it is not my area of expertise). Of course good requires a reasonable error metric, and I don't think that exists yet.