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 transcriptions for barely half the words spoken.
The researchers speculated that the systems had a common flaw: âinsufficient audio data from Black speakers when training the models.â
A startup called Speechmatics has developed a technique that appears to reduce this data gap.
The company announced last week that its software had âan overall accuracy of 82.8% for African American voicesâ 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%.
Speechmatics attributed much of its performance to a technique called self-supervised learning.
Training school
The advantage of self-supervised models is that they donât 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.
This helped Speechmatics increase its training data from around 30,000 hours of audio to around 1.1 million hours.
Will Williams, the companyâs VP of machine learning, told TNW that the approach improved the softwareâs performance across a variety of speech patterns:
What weâre looking to do is build scalable methods that let us attack a broad range of accents at once.
Learning like a child
One of the techniqueâs benefits was closing Speechmaticsâ age understanding gap.
Based on the open-source project Common Voice, the software had a 92% accuracy rate on childrenâs voices. The Google system, by comparison, had an accuracy of 83.4%.
Williams said enhancing the recognition of kidsâ voices was never a specific objective:
Weâre training on millions of hours of audio, and just like how a child learns, weâre exposing our learning systems to all this online audio⌠Inside those millions of hours, there will be childrenâs voices, so it will learn how to deal with them â but without them being labelled.
That doesnât mean that self-supervised learning alone can eliminate AI biases. Allison Koenecke, the lead author of the Stanford study, noted that other issues also need to be addressed:Â
We also strongly believe that achieving fair outcomes is as much a âpeople problemâ as a âdata problem.â That is, we hope that ASR [automatic speech recognition] developers themselves understand the need to be broadly inclusive.
Get the TNW newsletter
Get the most important tech news in your inbox each week.