The precondition for language-independent speaker ID method is the fact that the current methods of speaker identification in majority of cases do not afford to carry out examination without a linguist who masters the language of the speaker:
- Questionable record (record under examination) and sample record can contain speech on different languages. Neither analyst, nor prosecutor or judge can make a defendant to give speech samples in the language they need.
- There are languages with lots of dialects (e.g. Spanish, or Chinese). The dialects can differ a lot, so that in audio forensics practice we would have two different languages – one an analysis speaks and another one the defendant speaks (e.g. although both of them speaks Spanish but different dialects).
- A linguist who knows the language can have no idea of forensiudio
However, the creation of such a method when an analysis conducts ID examination of a person who speaks the language unknown to the examiner is possible and is based on the usage of instrumental methods.
It also affords to enlarge the number of ID examiners, they can be audio engineers who passed the training on “speech aspects” and linguists who can conduct spectral analysis
From the very beginning of the forensic speaker identification in Russia in 1949 [1] and up to now the comparative spectral analysis of the speech signal records is an important stage of speakers’ identity investigation for majority of experts. The evidence of the simple correspondence between the sound phonetic quality of the oral speech pattern and its sonogram is an excellent reason to transfer the problem of the aural distinguishing into the visual area. Classical works in “voice print” also give very stand-off and cautious proposals of using the spectral analysis of the speech in forensic speech investigation. Kuenzel [3] believes that spectral analysis is mainly only an excellent way to confirm and to objectify the thorough aural phonetic analysis of the speech samples under investigation. Hollien [4] is in a doubt at all that any instrumental means and spectral analysis in particular could help to distinguish different and equal speakers and acknowledges very reasonably this opinion by examples of some questionable sonograms. We do really believe that there is no difference in aural phonetic and instrumental spectral analysis in speaker’s ID process.
Like all known to us other approaches to forensic speaker ID our procedures have the strong scientific and statistical justification. We also have a positive real forensic practice of the methods reported here shows that they are sufficiently reliable, i.e. the probability of the mistake in the decision based on the given methods is compatible to the ones of the widely used forensic identification methods of other types of examinations: handwriting, some biological features, sometimes fingerprints, etc.
In this very short report it is impossible to explain and even to mention all essential aspects of the speech spectral analysis usage for forensic speaker ID. So here is the description of only one of the most important practical approaches, which has been useful for the most number of real cases.
By the means of the given method the work of the experienced expert on a complex of the analysis of speech signals "IKAR" with the program SIS of version 6.x , made by STC is proposed.
https://www.aes.org/e-lib/browse.cfm?elib=14416
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.
Learn more about the AES E-Library
Start a discussion about this paper!
