{"id":8833,"date":"2021-11-08T18:24:28","date_gmt":"2021-11-08T18:24:28","guid":{"rendered":"http:\/\/TheNextWeb=1371294"},"modified":"2021-11-08T18:24:28","modified_gmt":"2021-11-08T18:24:28","slug":"startup-harnesses-self-supervised-learning-to-tackle-speech-recognition-biases","status":"publish","type":"post","link":"https:\/\/www.londonchiropracter.com\/?p=8833","title":{"rendered":"Startup harnesses self-supervised learning to tackle speech recognition biases"},"content":{"rendered":"\n
<\/div>\n

Speech recognition systems struggle to understand African American Vernacular English (AAVE). In a 2020 study by Stanford University researchers<\/a>, the software performed so poorly for AAVE that some leading systems made correct transcriptions for barely half the words spoken.<\/p>\n

The researchers speculated that the systems had a common flaw: \u201cinsufficient audio data from Black speakers when training the models.\u201d<\/p>\n

A startup called Speechmatics has developed a technique that appears to reduce this data gap.<\/p>\n

The company announced last week<\/a> that its software had \u201can overall accuracy of 82.8% for African American voices\u201d based on datasets used in the Stanford study. In comparison, the systems developed by Google and Amazon both recorded an accuracy of only 68.6%.<\/span><\/p>\n

Speechmatics attributed much of its performance to a technique called self-supervised learning.<\/p>\n

Training school<\/h2>\n

The advantage of self-supervised models is that they don\u2019t require all their training data to be labeled by humans. As a result, they can enable AI systems to learn from a much larger pool of information.<\/p>\n

This helped Speechmatics increase its training data from around 30,000 hours of audio to around 1.1 million hours.<\/p>\n

Will Williams, the company\u2019s VP of machine learning, told TNW that the approach improved the software\u2019s performance across a variety of speech patterns:<\/p>\n

\n

What we\u2019re looking to do is build scalable methods that let us attack a broad range of accents at once.<\/p>\n<\/blockquote>\n

Learning like a child<\/h2>\n

One of the technique\u2019s benefits was closing Speechmatics\u2019 age understanding gap.<\/p>\n

Based on the open-source project <\/span>Common Voice<\/span><\/a>, <\/span>the software had a 92% accuracy rate on children\u2019s voices. The Google system, by comparison, had an accuracy of 83.4%.<\/span><\/p>\n

Williams said enhancing the recognition of kids\u2019 voices was never a specific objective:<\/p>\n

\n

We\u2019re training on millions of hours of audio, and just like how a child learns, we\u2019re exposing our learning systems to all this online audio\u2026 Inside those millions of hours, there will be children\u2019s voices, so it will learn how to deal with them \u2014 but without them being labelled.<\/p>\n<\/blockquote>\n

That doesn\u2019t mean that self-supervised learning alone can eliminate AI biases. Allison Koenecke<\/a>, the lead author of the Stanford study, noted that other issues also need to be addressed: <\/span><\/p>\n

\n

We also strongly believe that achieving fair outcomes is as much a \u2018people problem\u2019 as a \u2018data problem.\u2019 That is, we hope that ASR [automatic speech recognition] developers themselves understand the need to be broadly inclusive.<\/p>\n<\/blockquote>\n

Nonetheless, the performance of Speechmatics suggests that self-supervised learning can at least mitigate dataset biases.<\/span><\/p>\n

Source<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Speech recognition systems struggle to understand African American Vernacular English (AAVE). In a 2020 study by Stanford University researchers, the software performed so poorly for AAVE that some leading systems made correct…<\/p>\n","protected":false},"author":1,"featured_media":8834,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"_links":{"self":[{"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=\/wp\/v2\/posts\/8833"}],"collection":[{"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8833"}],"version-history":[{"count":0,"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=\/wp\/v2\/posts\/8833\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=\/wp\/v2\/media\/8834"}],"wp:attachment":[{"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8833"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8833"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.londonchiropracter.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8833"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}