We've got rhythm
THESE days a lot of crime is planned, executed, and sometimes gloated over, via the medium of the mobile phone. So a new field of forensic phonetics has arisen, which tries to identify people from their voices. But the technology is fallible and criminals often have the last, unidentifiable, laugh. But the International Crime Science Conference, held recently in London, heard that better techniques are on the way.
Voices acquire idiosyncratic features from a number of sources such as parents, co-workers and school friends. Accents and vocal mannerisms can help identify a person but these can also be disguised. Something that is harder to cover up is the stamp that the anatomy of a person’s vocal tract leaves. The smaller the larynx, for example, the higher the frequency at which the vocal cords vibrate, and the higher-pitched the voice.
This “fundamental frequency” is fairly stable across a person’s speech, and it is one of the characteristics voice experts use to try to match, say, a taped message from a terrorist or hoax call to a recording of a suspect. But fundamental frequencies, too, can be disguised.
Volker Dellwo, a phonetic scientist from University College London, conducted a small study in which he asked an actor to read 25 sentences in either his natural voice or with an accent. Dr Dellwo found that the actor’s fundamental frequency went from 120 hertz to 170 hertz when disguised. Furthermore, phonetic experts found it difficult to say in many cases if the same person was reading the sentence.
Dr Dellwo thinks it would be better to focus on rhythmic aspects of speech, such as the way a person lingers over vowels. He has found that the percentage of speech that is vocalic, as opposed to consonantal, is not only highly idiosyncratic and stable regardless how fast a person is speaking, but also is not perceived by the ear, making it hard to disguise. In the same study he found that this aspect of the actor’s speech did not change much, whether he read in a normal or disguised voice.
This study needs to be replicated with many more subjects and types of speech, but the results hint that speech rhythm could be profitably tapped by voice analysts. One snag is that, at present, the visual representation of the voice has to be tediously annotated manually for its vowel and consonant content.
However, a slightly different approach, measuring the proportion of speech time during which the vocal cords vibrate, could be automated—since software for extracting that feature already exists. If this proves viable it could become useful in police work, prosecutions and courtrooms. Scientists may yet have the last laugh.
driver from www.economist.com