Robustness of Long Time Measures of Fundamental Frequency

• In many speech technology based applications as well as in forensic phonetics it is desirable to obtain reliable estimates of a speaker’s fundamental frequency. We would like the measures to be accurate and reliable enough in order to be used meaningfully as a parameter in speaker identification or verification. Under optimal conditions such as when high quality studio recordings and normal speech styles are used this is often possible. In real life applications such conditions are the exception rather than the rule. The study presented here reports the result from an investigation where different measures were tested on speech material that varied with respect to speaking style, vocal effort and recording quality. Based on the results from these tests we would like to suggest a measure we call the alternative fundamental frequency baseline as the measure that is most robust with respect to the above-mentioned sources of variation. Index Terms: speaker recognition, speaker identification, fundamental frequency, F0.

Ämnesord

HUMANIORA  -- Språk och litteratur -- Studier av enskilda språk (hsv//swe)

HUMANITIES  -- Languages and Literature -- Specific Languages (hsv//eng)

Nyckelord

Speaker identification

acoustic phonetics

forensic phonetics

speech technology

Publikations- och innehållstyp

ref (ämneskategori)

kon (ämneskategori)