Software Technologies Working Group (GTTS)
|
Department (s) |
||
|
Knowledge area |
||
| PI: Amparo Varona Fernández | Co-PI: | |
|
Members Germán Bordel García, Mikel Peñagarikano Badiola, Luis Javier Rodríguez Fuentes |
||
|
Keywords Machine Learning, Deep Learning, Spoken language Recognition, Speaker Recognition, Spoken Term Detection, Automatic Speech Recognition, Speech-to-Text Alignment, Automatic Subtitling, Design and Collection of Speech Databases |
||
|
Description Since about 2005, GTTS research activities have focused on speech processing, applying Machine Learning (ML) and more recently Deep Learning (DL) techniques. Since 2010, GTTS is in charge of the automatic subtitling of videos of the Basque Parliament plenary sessions, based on the official minutes. GTTS has developed algorithmic improvements to spoken term detection (STD) and query-by-example STD systems, based on ML and DL approaches. GTTS has also made several contributions to the fields of spoken language and speaker verification, and has participated in international benchmark evaluations: NIST 2007, 2009, 2011, 2015, 2017 Language Recognition Evaluation and NIST 2008, 2010, 2012, 2019 Speaker Recognition Evaluation. In 2019, GTTS obtained, together with the Aholab group of the Bilbao Engineering School, the accreditation as Research Group A (Aholab-GTTS: Audio, Speech and Language Processing Group) in the general call for research groups of the Basque Government. |
||
|
Research lines
|
||
|
Equipment The group manages 6 computing servers with capacities of between 9000 and 47000 MFlops, RAM between 24 and 128 GB, all of them under CentOS and one of them with GPUs, as well as 2 NAS disk servers (main unit: 40TB, backup unit: 80TB). The group has access to a fair amount of speech databases, some of them of its own and others obtained through NIST evaluations or from an annual subscription to the LDC (Linguistic Data Consortium). |
||
|
Website link |
Contact |
|