Speechmatics outperforms Google in voice recognition

Thank you for reading the news about technology: Speechmatics beats Google in voice recognition and now with the details

Cairo – Samia Sayed – A British speech recognition startup said its voice recognition technology was better than the performance of major tech companies such as Google and Amazon at understanding the voices of black people, according to a report by the Arab Gateway for Technical News.

Speechmatics said its system has an overall accuracy rate of 83 percent for African American votes. This is higher than Microsoft (73%), Amazon (69%), Google (69%), and Apple (55%), according to research published by Stanford University in 2020.

The Stanford University research compared the results of major tech companies in how accurate speech recognition software is in understanding African Americans. Moreover, Amazon, Google, Microsoft and Apple systems made nearly twice as many errors when interpreting words spoken by African Americans compared to whites, according to researchers at Stanford University.

Speechmatics says its system misrecognizes words from black voices 17% of the time, compared to 31% for Google and Amazon.

“It is critical to study and improve fairness in speech-to-text systems because of the potential for disparate harms to individuals across downstream sectors ranging from healthcare to criminal justice and voice recognition technology has become an integral part of everyday life,” said Alison Koenicki, lead author of the Stanford study. This is thanks to the proliferation of virtual assistants via smart devices such as phones and speakers.

Apple pioneered the use of voice-activated software on mobile devices through its digital assistant Siri, while Amazon was one of the first to bring speech recognition into the home with its Echo speakers and Alexa assistant. increasingly concerned about bias in the algorithms that support these speech recognition services.

Experts say many voice-recognition programs are trained on limited sets of data, which makes them less effective, and it is related to the quality of the data in the training sets.

And there’s been racial bias, gender bias, and regional dialect bias in speech recognition technology for a long time, and this technology doesn’t work the same way for everyone yet.

Speechmatics says it trained its AI with unlabeled data from social media and podcasts, to help it learn different aspects of speech including dialect and language. Moreover, the company said its technology is trained for 1.1 million hours of audio.

Speechmatics described the development as a breakthrough, and hopes that other tech companies will become more transparent about efforts to reduce bias in artificial intelligence, and as a result tech giants ramped up their investments in speech recognition recently, and Microsoft agreed to acquire software company Nuance Communications for $16 billion. dollars in April.