GTM-UVigo Systems for Albayzin 2016 Search on Speech Evaluation

TitleGTM-UVigo Systems for Albayzin 2016 Search on Speech Evaluation
Publication TypeConference Proceedings
Year of Publication2016
AuthorsLópez Otero, P, Docío Fernández, L, García-Mateo, C
Conference NameIberspeech 2016
AbstractThis paper describes the systems developed by the GTM-UVigo team for the Albayzin 2016 Search on Speech evaluation. The system for the spoken term detection task consists in a large vocabulary continuous speech recognition approach which features a strategy for out-of-vocabulary term detection: string search of the phonetic transcription of the search terms is performed within the most likely sequence of phonemes output by the speech recogniser. For the query-by-example spoken term detection task, a language-independent approach is proposed, which is a combination of three systems based on dynamic time warping search that differ in the speech representation strategy: one relies on phoneme posteriorgrams obtained from phone models in English; another one represents speech by means of Gaussian posteriorgrams; and the remaining one represents the audio documents using acoustic features. The phoneme posteriorgram and acoustic feature representations implement a strategy to select the most relevant phoneme units and features, respectively.
Citation Key606