Behavox announces launch of voice biometrics
02 July 2019 16:52 GMT

Behavox, an enterprise People Analytics software company, announced today the launch of voice biometrics within Behavox Voice and new Chinese language capabilities, with an initial focus on Mandarin and Cantonese.

In addition to voice biometrics, new capability highlights include Chinese speech transcription, keyword spotting functionality and distinct language identification from audio.

Behavox Voice launched in 2018 and boasts a suite of proprietary capabilities that enables organizations to process and analyze voice recordings through a variety of features, like multilingual functionality. In March of this year, Behavox announced new Spanish voice capabilities following client feedback – and today’s launch of Voice Biometrics, in addition to Mandarin and Cantonese functionality, illustrates just how quickly Behavox is progressing in its expansion into multiple dialects.

“With nearly 1.2 billion Chinese language speakers worldwide – equating to 16% of the world’s population – we knew that Mandarin and Cantonese had to be the next languages launched within Behavox Voice,” said Erkin Adylov, Founder and CEO of Behavox. “As a global artificial intelligence firm, it is imperative that we develop these languages for our client base, especially with a great deal of clients concentrated across Asia. What’s more, new voice biometrics’ capabilities allow clients to identify speakers, even in noisy environments, with as little as 5 minutes of speech.”

Voice biometrics touts the ability to transcribe an employee’s voice in a noisy environment with unparalleled transcription accuracy of over 94%. The voice biometrics platform can identify a speaker based on the person’s vocal footprint with ease and accuracy – advanced machine learning models analyze unique features present in the pitch and tone of the human voice to accurately identify individuals in phone recordings. This allows individual’s phone calls to be attributed to him or her even if an unknown phone number(s) is used, which is a significant feature following GDPR and other regulations put in place by the SEC and various regulators.

Behavox’s dedicated Voice Lab recognized a few glaring problems when they embarked on the integration of Mandarin and Cantonese into Behavox Voice – both languages are difficult to distinguish from audio, speakers frequently switch between the two languages, there is a general lack of data available for Cantonese, and finally, Chinese characters are logograms, representing words or phrases rather than sounds. This presents a huge challenge to engineers working to integrate variants of this language into systems.

Behavox’s solution involves a single, multilingual model and shared lexicon (a single list of words), which allows the model to transcribe both languages without error – allowing a Chinese speaker to switch freely back and forth between Mandarin and Cantonese.

A novel feature of the new Chinese language capabilities within Behavox Voice is the model’s ability to recognize characters instead of words (like in English and Spanish). With over 4,142 hieroglyphs in Cantonese and Mandarin, this is a significant nuance that Behavox programmers built into the system.

Industry Events