 |
| Speechmatics outperforms Google in speech recognition |
A UK speech recognition startup said its speech recognition technology is outperforming big tech companies like Google and Amazon at understanding black voices.
Speechmatics claims that their system has an overall accuracy rate of 83% for African American votes.
According to a study published by Stanford University in 2020, this is more than Microsoft (73%), Amazon (69%), Google (69%) and Apple (55%).
The Stanford University study compared results from large tech companies on how accurately their speech recognition software understands African Americans.
Importantly, according to researchers at Stanford University, Amazon, Google, Microsoft, and Apple systems make twice as many mistakes as whites when interpreting what African Americans say.
Speechmatics said its system was bad at detecting black noises 17% of the time, compared to 31% at Google and Amazon.
Lead author of the study at Stanford University, Alison Konecki, said, “Research and improvement of equity in text-to-speech systems is critical because areas of criminal justice healthcare can lead to personal failure and relative harm.
Speech recognition technology has become an indispensable part of everyday life thanks to the proliferation of virtual assistants through smart devices such as phones and speakers.
Apple is the first to use voice-activated software on mobile devices with the Siri digital assistant.
Amazon is one of the first companies to bring voice recognition into the home with its Echo speakers and Alexa assistant.
Distortions in speech recognition technology
Researchers are increasingly concerned about inconsistencies in the algorithms that support these speech recognition services.
Experts say many speech recognition programs are trained on a limited set of data, so their effectiveness is low.
It relates to the quality of the data in the training set. Speech recognition technologies for racial, gender, and regional biases have been around for a long time. This technology works differently for everyone.
Speechmatics uses unclassified data from social media and podcasts to train its AI. This will help him learn different aspects of phonetics, including dialects and languages.
Most importantly, the company's technology went through 1.1 million hours of voice training. Theology calls this development a breakthrough.
It hopes other tech companies will become more transparent in their efforts to reduce bias from AI.
Hence, tech giants have recently increased their investment in speech recognition. Last April, Microsoft agreed to acquire software company Nuance Communications for $16 billion.