Authors: Volokhov V., Lavrentyeva G., Novoselov S., Matveev Y.
Description: the project is related to the development of labs for the ITMO Speaker Recognition Course.
Keywords: voice biometrics, speaker recognition, speaker verification, speaker identification, acoustic features, speech activity detector, machine learning, speaker embedding extractor, deep neural network, decision theory, domain adaptation and calibration.
Datasets: the main databases for performing of labs is VoxCeleb corpus.
Content: the repository contains materials (now only in russian language) for self-performing five labs. The titles of the labs are listed below.
- Lab work 1. Informative features of speech signals: feature extraction (link).
- Lab work 2. Voice activity detector training (link).
- Lab work 3. Creating and comparing speaker models (link).
- Lab work 4. Decision criteria and quality metrics (link).
- Lab work 5. Adaptation and calibration of speaker recognition system (link).
Some ideas for creating of labs were borrowed here (training of voice activity detector model), here (training and testing of speaker embedding extractor) and here (training of calibration model for voice biometrics system).
A published version of these labs (now only in russian language) can be found here. Publication date: 05/24/2022.
A latest updated version of these labs (now only in russian language) can be found here. Publication date: 05/24/2022.