Researchers develop new method for ensuring quality in voice biometrics data collection
A new artificial intelligence-based method for collecting voice biometrics by ensuring the quality of automatic voice recordings has been developed by researchers from HSE University and Nizhny Novgorod State Linguistic University (LUNN). The method involves an algorithm resistant to noise of 10dB or higher which can operate in real-time, and could have significant implications for speech recognition.
The researcher’s findings are presented in a new paper published in Measurement Techniques titled “A Method for Measuring the Pitch Frequency of Speech Signals for the Systems of Acoustic Speech Analysis.” The poor quality of voice reference templates, usually due to ambient noise, is a constraining factor for the widespread adoption of voice identification systems, according to the announcement.
The method proposed by Professor Andrey Savchenko of HSE University and Professor Vladimir Savchenko of LUNN can reduce the error rate of voice identification systems to 2 percent at a signal-to-noise ration of 10dB or higher, they say.
The proposal also includes the use of an algorithm to split speech into short frames, and the range of pitch frequencies in each is compared to identify good quality recordings.
A major Russion bank is reported to have interest in the technology, and have provided recordings from its voice database for initial testing.
The global market for speech and voice recognition was recently forecast to grow beyond $28 billion by 2026.