Automatic Classification of Languages
Abstract
Studies were made of language classification algorithms which were based on the recurrence frequencies of sequences of phoneme-like sound segments. Sequences of length 2, 3, 4 and 5 were considered. The sequence recurrences comprised acceptances of acoustic hypotheses, which in turn are based on a time- frequency scanning error measure, and on occurrence time relationships. Classification performance was estimated using 50 independent test speakers of five languages. Individual language accuracies ranged from 17 to 90 percent, with the overall five-language accuracy being 70 percent.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1975
- Accession Number
- ADB008708
Entities
People
- George R. Doddington
- R. G. Leonard
Organizations
- Texas Instruments