VOICE QUALITY AND FORENSIC SPEAKER IDENTIFICATION

When phoneticians compare forensic speech samples they often remark in their reports on a "similarity of voice quality". Likewise, when earwitnesses are asked to describe a voice they have heard, they will normally comment on the accent, if they are able to, and additionally describe what they heard as an "X voice" where "X" is a term such as "rough" or "resonant" that can be seen as an informal label of voice quality. In this talk I will examine these two main categories of forensic speaker identification - by phonetic experts and by earwitnesses - with reference to the notion of voice quality. I will take voice quality in the broad sense discussed by Laver in his The Phonetic Description of Voice Quality (CUP, 1980), that is, as covering supralaryngeal as well as laryngeal characteristics which emerge cumulatively from a person's speech. In speaker comparison by phonetic experts the emphasis in acoustic analysis tends to be on segmental properties, or on pitch-related long-term features. I will give some examples of how speakers can be differentiated in this way, and touch on how the dynamics of formants in transitional parts of the speech signal may provide the nearest we have to a speaker's "signature". Beyond segmental analysis, however, I will show that an analysis using the long- term distributions of formant frequencies can capture information relating to Lover's supralaryngeal voice quality categories. Given the availability of Laver's comprehensive framework for the impressionistic analysis of voice quality we might ask why, in the auditory strand of their forensic analyses, phoneticians have made little use of systematic voice quality description, and I will explain why I think that is. As regards earwitness evidence I will focus on the description of voices by earwitnesses, and on the use of voice parades. I will ask whether an , earwitness 's description of a voice might be improved if questioning of the witness were informed and structured by knowledge of a framework for voice quality description. And in creating a voice parade, I will show how pre-tests are used to ensure that the parade is fair, including one where experimental subjects are, in effect, asked to rate the similarity in voice quality between all pairs of samples to be used in the parade. This is to ensure that the suspect is not an outlier. Finally I will preview a project which will investigate the effect of the telephone on such similarity judgments.