Findings: Cats' chorus to choir of angels

If you sound like a cat being strangled when singing, help could soon be at hand, writes Natasha Gilbert.

Mark Smith, head of Purdue University's School of Electrical and Computer Engineering, in the US, and graduate student Matthew Lee of the Georgia Institute of Technology claim to be able to make even the poorest of voices sound good with the aid of a computer model that analyses and deconstructs a person's voice and then synthesises a digitally enhanced version of it.

"Our programme can transform a poor singing voice into a great singing voice by altering its pitch, duration and vibrato," Professor Smith said.

In the 1980s, Professor Smith and Brian George, his doctoral student at the Georgia Institute of Technology, engineered a system called the sinusoidal model. This breaks down a human singing voice into components (or sine wave segments) with different frequencies, amplitudes and phases.

Professor Smith and Mr Lee developed a method of modifying these voice components and then reconstructing the singing while maintaining the naturalness and identity of the original voice.

"The way the components are modified leads to the improvements that we hear," Professor Smith told The THES . "We have used several good singers as examples to govern how we do our modifications."

While Professor Smith and Mr Lee have improved the quality of singing in voice samples, they have yet to improve a whole song. The researchers hope to apply the sinusoidal model to improve the quality of text-to-speech programmes, in which words typed into a computer are automatically converted into spoken languages, and programmes for the hearing impaired.

But Professor Smith said that his ultimate aim was to cut an album.

This project was presented at the 145th meeting of the Acoustical Society of America, Nashville, Tennessee, this week.