Likelihood ratios are a Bayesian statistical tool used in forensic voice comparison to quantify how much more probable the observed speech evidence is under the same‑speaker hypothesis than under the different‑speaker hypothesis. They provide a scale of similarity versus typicality for each analysed acoustic feature, allowing the strength of evidence for or against speaker identity to be expressed either as a numeric estimate or as a verbal statement such as “the evidence is much more probable if the same‑speaker hypothesis were true.” The overall conclusion of a forensic voice comparison is derived from the combined likelihood‑ratio assessments across multiple independent speaker characteristics. When the acoustic data are insufficiently clear—e.g., similarity is neither high nor low, features give conflicting results, or the quantity/quality of material is too low—the comparison may remain inconclusive, and evaluators may resort to subjective numeric estimates if formal calculations cannot be performed.
Definition drawn from The Routledge Handbook of Forensic Linguistics (Coulthard, May & Sousa-Silva, eds., 2nd ed.). Extracted text: /Volumes/mu-not/projects/zodiac/books/routledge_handbook_of_forensic_linguistics/routledge_handbook_of_forensic_linguistics.txt.
Bears on: authorship & stylometry · the letters · sources