Speaker verification is the task of estimating how likely it is that two speech recordings come from the same speaker, while speaker identification tries to match a speech recording with one of a greater number of candidates. Both are important forensic linguistics/phonetics tasks and crucial for law enforcement.

Google and other sources told me that a range of phonetic features such as pitch range, long term formant frequencies, speech rate and articulation rate are used in these tasks. Unfortunately all sources I found were fairly general. My question is: Which of these (and other phonetic features) are considered most reliable and are most commonly used? If it's important for answering the question, I'm mostly (but not exclusively) interested in English here. Also, answers should ideally refer to (academic) sources.

  • Well, they have spent the money and done the research and they know whatever they've found out, and they're not telling because it's making them money. Trade secrets. Anyone else is also at liberty to do the research themself, for whatever purposes they have in mind. Bear in mind that there are a lot of different purposes for developing this technology, and they don't all have the same standards of success, so they won't all be designed the same way or use the same features or algorithms (which are also proprietary, natch).
    – jlawler
    Aug 18, 2014 at 13:49


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.