atlanTTic, represented by GTM-UVigo team, has attended the conference IberSpeech 2014, where it has presented two papers and it has won the “Search-on-speech” evaluation.
The paper, carried out by Paula López, Laura Docío and Carmen García, deals with the use of iVector speech representation to perform continuous emotion recognition.
The other paper was conducted by Marta Martínez and Antonio Cardenal and it involves the study and analysis of basic strategies for keyword spotting (search for key words in audio content) in a database in Galician language. It was also presented the work carried out in the project SpeechTech4All, in particular a graphical interface that serves as a demonstration of the technologies developed by the different groups which form part of the project SpeechTech4All: University of Vigo, University of the Basque Country and the Polytechnic University of Catalonia.
IberSpeech 2014 is a conference on several topics related to speech technologies, with special interest on the languages spoken in the Iberian Peninsula. It is organised by the Spanish Thematic Network on Speech Technologies (RTTH) and the ISCA Special Interest Group on Iberian Languages (SIG-IL). This conference is being held every two years since 2000; the latest editions were held in Vigo in 2010, under the name FALA 2010, and in Madrid, where the name IberSpeech was adopted.
During the conference the researchers of the Multimedia Technology Group (GTM) from the centre atlanTTic were present in two of the “Albayzin evaluations”, winning the “search-on-speech” evaluation and being second on audio segmentation. The “search-on-speech” evaluation consisted in the search for keywords in audio contents in two different modes: in one of them the key word (or words) is given as text, whereas, in the other, an acoustic example of the word is given (that is to say, a voice file in which someone utters the word). The GTM-Uvigo team (composed by Marta Martínez, Paula López, Rocío Varela, Laura Docío, Antonio Cardenal and Carmen García) was the one who developed the best system for all modes.
Other topics, such as speech recognition, speech synthesis, language and speaker characterisation, speech and language technologies in different fields of application, were discussed at the conference.
In addition, there was a special session on Albayzín evaluations and another on thesis, demos and projects.