We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
ALINEACIÓN FORZADA SIN ENTRENAMIENTO PARA LA ANOTACIÓN AUTOMÁTICA DE CORPUS ORALES DE LAS LENGUAS INDÍGENAS DE COSTA RICA.
- Authors
Coto-Solano, Rolando; Flores Solórzano, Sofía
- Abstract
Forced alignment provides drastic savings in time when aligning speech recordings. This is particularly useful for Indigenous languages, which lack tagged corpora and resources for their computational study. In this article we present a method for the alignment of Bribri, Cabecar and Malecu recordings using acoustic models trained for English and French. We used the FAVE-align and EasyAlign to produce Praat TextGrids, and obtained error rates of 2~3 milliseconds when marking the center of Bribri and Malecu words (8~13% of average word duration), and of 7 milliseconds when marking Cabécar words (37% of average word duration). Phoneme alignment also showed an adequate performance: An average of 40% of Bribri and Malecu phonemes were aligned with an error of 1 millisecond or less; while 24% of Cabécar phonemes had the same error rate. The lower performance when aligning Cabécar might have been caused by a higher level of environmental noise in the recording used. These forced alignment systems can assist in the study of the Indigenous languages of Costa Rica, in particular in the generation of aligned corpora for phonetic study and for the training of acoustic and speech recognition models.
- Publication
Káñina. Revista de Artes y Letras de la Universidad de Costa Rica, 2016, Vol 40, Issue 4, p175
- ISSN
0378-0473
- Publication type
Article