Non-contemporary speech samples: auditory detectability of an 11 year delay and itseffect on automatic speaker identification


  • Hermann J. Künzel



, non-contemporary speech, automatic speaker identification, forensic phonetics


The need to compare non-contemporary speech samples is a common issue in forensic speaker recognition. Even if the size of the delay is known a crucial question has still to be answered in every single case: Is it possible that physiological aging or other factors have caused alterations of parameters of voice, speech and language to such a degree that the material basis for identification by human listeners and/or by automatic systems may no longer be regarded as adequate when a new identification task is due? Some studies on auditory speaker recognition suggest that, other variables being equal, a time lag of up to 6 years does not seem to pose a problem. To this author’s knowledge, the question has not yet been directed towards automatic identification systems. The present study uses speech data from ten male speakers recorded at intervals of 11 years and analyses the effect of the delay in terms of (a) the ability of different groups of listeners to detect it, and (b) its influence on the performance of an advanced automatic speaker identification system for forensic applications. Using contemporary samples for both ‘known’ and ‘unknown’ models as a benchmark the automatic speaker identification system identified all speakers correctly with LRs between 102 and over 108. In the non-contemporary condition, where the older samples (i.e. ‘younger’ speakers) were used for the construction of models for ‘known’ and the samples recorded 11 years later for ‘unknown’ speakers, LRs of nine speakers remained unchanged or dropped only slightly. The LR for one speaker dropped sharply. It seems that this is the only case in which vocal aging, but perhaps also other time-related factors, may have played a rôle. The main conclusions of these experiments are that for most male speakers a delay of the size of a decade between voice samples will not pose a problem to either auditory or machine-based speaker identification.

Author Biography

Hermann J. Künzel

HERMANN J. KÜNZEL (MA, PhD, Kiel) has published widely on aspects of general and forensic phonetics. He was Head of the Speaker Identification and Tape Authentication Section of the Bundeskriminalamt in Wiesbaden from 1985 to 1999 and is now Professor of Phonetics at Marburg University. He has been working as an expert in speaker identification, speaker profiling, voice line-ups and non-speech related acoustic investigations (e.g. aircraft mishaps) for courts and government institutions throughout Germany and worldwide. Besides forensic phonetics, his academic interests focus on physiological and pathological aspects of phonetics.



How to Cite

Künzel, H. J. (2007). Non-contemporary speech samples: auditory detectability of an 11 year delay and itseffect on automatic speaker identification. International Journal of Speech, Language and the Law, 14(1), 109–136.




Most read articles by the same author(s)