Building a Speaker Clustering Tool by Employing Speaker Recognition Techniques

LOCATION:  TEL, Auditorium 3 (20th floor), Ernst-Reuter-Platz 7, 10587 Berlin

Date/Time: 23.10.2017, 14:15-15:00

SPEAKER: Renzo Verastegui  (TU Berlin)

ABSTRACT:

The analysis of voice similarity is useful in many research fields, such as speaker diarization, speech synthesis, and voice casting. Applications that measure the similarity of voice are extremely expensive to administer and also require a significant amount of expert listening. For these reasons, it is of main interest to be able to automate the detection of speech similarities in voice segments.
The aim of this Master Thesis is to build a speech clustering detection tool based
on speaker recognition state-of-the-art techniques: i-vectors and deep neural networks, among others. Finally, the relevance of the generated clusters is determined by a comparison to a previously conducted study of subjective similarity.