Bulletin

Issue: May 2006

Research in Action

Analysing rhythms and melody of speech

Why does artificial speech emanating from a computer sound so dull and flat, so robotic – so unhuman?

It is, according to La Trobe University phonetician Dr Marija Tabain, because the current generation of computers cannot control how consonants and vowels interact with the rhythm and melody of speech.

But now there is hope that the voice of speaking computers of the future may sound more like ours.

And, more importantly, there is hope that speech therapists working to improve the speech of brain damaged people may be able to devise more effective therapies.

Dr Tabain was recently awarded a $150,000 ARC Discovery Grant over three years to research into an area of speech called prosody – the science of the rhythm and pitch of speech.

While emphasising that her project is basic research, she says her results will feed into future outcomes such as improving computer articulation and speech therapy.
To do this Dr Tabain will work with acoustic recordings, and then use sophisticated computer techniques to analyse the sounds.

In addition she will use other techniques, such as electromagnetic articulography which tracks the movement of the tongue, lips and jaw, and electropalatography, which measures contact between the tongue and the palate.

She will work with voice recordings in three languages: English, French and the Australian Aboriginal language, Arrernte, spoken in and around Alice Springs.

Dr Tabain said that the structure of different languages influenced the interaction of consonants and vowels with speech melody and rhythm.

For example, French speakers tend to control their tongue and lip movements for ‘i’ and ‘u’ much more carefully than do English speakers. The French vowels are relatively less affected by the rhythm and melody of the language than are the English vowels. This is because French has a special vowel found in words like ‘tu’ (‘you’), which is easily confused for ‘i’ or ‘u’ in some situations.

‘And of course, almost nothing is known about how consonants and vowels interact with the rhythm and melody of speech in Australian Aboriginal languages,’ said Dr Tabain.

‘The importance of rhythm and melody to the naturalness of speech is seen in some speakers who have suffered traumatic brain injury. They perform well when asked to utter a short word, but struggle when asked to produce a longer string of sounds. It’s as if producing the consonants and vowels is hard enough, but adding rhythm and melody is just too much.

‘A better understanding of the interaction between speech sounds and prosodic structure will lead to improvements in the treatment of speech disorders, and in the quality of human-machine communication.’