789 views
Russian voice biometrics technology demonstrates outstanding results
“High-quality voice identification of a person helps to improve business and public services, making our lives easier. High-end speech technologies help to create the best interactive assistants, streamline the work of call centers, sales offices and service points. Speech analytics helps to draw conclusions about customer satisfaction and the dialogue quality, and therefore continuously improve the user experience,” commented STC Group CEO Dmitry Dyrmovsky.
The technology of the STC Group showed an outstanding result in the NIST SRE21 (Speaker Recognition Evaluation) competition.
The competition offered several tasks to solve:
• Speaker detection based on different audio sources: phone calls (conversational telephone speech, CTS), audio from video (AfV). For this purpose, an algorithm for a person identification by voice was applied.
• Speaker detection based on audio and video from various sources: phone calls (CTS), audio from video (AfV) and just video. For this purpose, a combination of algorithms for a person identification by voice and by face was used.
The STC scientific team was one of the first who solved the tasks of person recognition successfully using a combination of transformer-type neural network architectures, which is popular in the tasks such as computer vision, natural language understanding, and wav2vec used in speech recognition tasks. This approach made it possible to achieve a low level of error in verifying a person by voice.
STC Group of Companies (part of the Sber ecosystem) is a global developer of products and solutions based on conversational artificial intelligence, machine learning and computer vision. With its 30 years’ experience, STC offers expertise in speech technology, facial and voice biometrics.