Friday, September 14, 2007

Genetic algorithm tunes up public speakers

Genetic algorithm tunes up public speakers

  • 19:00 17 July 2002
  • Exclusive from New Scientist Print Edition. Subscribe and get 4 free issues.
A new voice processing system could help you "evolve" your voice so that it sounds the way you want it to when you speak through a microphone. The technology could turn weak and weedy voices into the expressive and clear ones needed for effective public speaking.

For some, having the right voice can make the difference between success and failure. Britain's ex-Prime Minister Margaret Thatcher famously had voice training to lower the pitch of her voice, making her sound more authoritative.

If you want to command respect, having a strong, authoritative voice helps, says Yuji Sato at Hosei University in Tokyo. So to help people develop it, he's devised a voice processing system that makes people sound more "joyful, calm or manly" - even the women.

His system uses a genetic algorithm to "evolve" an improved voice, he told the Genetic and Evolutionary Computing conference in New York last week.

Voice chromosomes

Starting with sample sentences, the algorithm analyses the voice signal to work out which aspects of it need to be enhanced or suppressed to produce the required effect.

To do this, it randomly creates a series of "voice chromosomes" representing ways in which the voice could be modified. Each chromosome is made up of three genes, corresponding to changes in voice pitch, volume and speed.

Individual chromosomes are applied to the waveform of the recorded speech, and the results played to observers who rank them according to the required criteria.

Rich and authoritative

After the group has ranked the new voices, the system takes the most successful voice chromosomes and swaps genes between them to produce a new generation of chromosomes. In addition, the software can randomly tweak certain genes, producing potentially interesting variations in the calmness, manliness or joyfulness of the voice.

Within a few generations, Sato's system has been able to create speech chromosomes that turn a feeble voice into a clear, rich authoritative one.

Unfortunately, Sato's system cannot yet reproduce the converted voice in real time, but he is working on it. However, even the current version should be good enough to make recorded TV soundtracks and computer game narratives more interesting.

No comments: