Google’s AI for Social Good team has been committed to applying artificial intelligence (AI) to solve various pressing issues, ranging from healthcare to climate change. One such initiative is Project Euphonia, which seeks to improve automatic speech recognition (ASR) technology for individuals with disordered speech patterns, offering them better communication tools and increased accessibility.
Traditional ASR systems have a significantly lower word error rate (WER) for individuals with typical speech. For those with disordered speech – including stuttering, dysarthria, and apraxia – high WERs make ASR far less effective. This disparity highlights the necessity for ASR models that cater to a wider range of speech abilities, ensuring inclusivity and accessibility for all users.
A key factor in improving ASR for disordered speech is collecting extensive and diverse samples. Over 1,000 hours of disordered speech samples were gathered from more than 1,000 participants, enabling the crafting of personalized ASR models. By tailoring these models, the performance gap for those with speech disorders was considerably reduced. Additionally, layer freezing techniques were implemented, streamlining the process by requiring only 3-4 minutes of training speech.
Collaborating with Google’s Speech team, the AI for Social Good team introduced Project Relate, which allows users with atypical speech patterns to create their own customized ASR models for enhanced communication. Potential applications of this technology include digital assistant tools and dictation apps, opening new avenues for people with speech disorders.
Despite the progress made through personalization, issues persist. Personalized models necessitate extensive recording sessions and often struggle to perform well in unscripted conversations. SI-ASR offers a potential solution by eliminating the need for user-specific training.
Developing SI-ASR models hinges on the quality of the training data. The Euphonia corpus was subdivided into train, validation, and test sets, forming the Prompted Speech dataset. Constructing representative dataset splits that encompass diverse speech impairments and underlying etiologies is vital for successful modeling.
By refining the USM to accommodate disordered speech without requiring personalization, the AI for Social Good team aims to make digital assistant technologies, dictation apps, and daily conversations more inclusive. The application of these advancements promises a more accessible future for users with speech disorders.
Improving ASR for people with disordered speech is essential in fostering an inclusive digital landscape. The progress made through personalization, SI-ASR implementation, and Project Relate demonstrates the potential for positive social impact. Ongoing research and development in this area will enable a more equitable and accessible future for all.
Up until working with Casey, we had only had poor to mediocre experiences outsourcing work to agencies. Casey & the team at CJ&CO are the exception to the rule.
Communication was beyond great, his understanding of our vision was phenomenal, and instead of needing babysitting like the other agencies we worked with, he was not only completely dependable but also gave us sound suggestions on how to get better results, at the risk of us not needing him for the initial job we requested (absolute gem).
This has truly been the first time we worked with someone outside of our business that quickly grasped our vision, and that I could completely forget about and would still deliver above expectations.
I honestly can't wait to work in many more projects together!
![The ‘Giveaway Piggy Back Scam’ In Full Swing [2022]](https://www.cjco.com.au/wp-content/uploads/pexels-nataliya-vaitkevich-7172791-1-scaled-2-683x1024.jpg)
Disclaimer
*The information this blog provides is for general informational purposes only and is not intended as financial or professional advice. The information may not reflect current developments and may be changed or updated without notice. Any opinions expressed on this blog are the author’s own and do not necessarily reflect the views of the author’s employer or any other organization. You should not act or rely on any information contained in this blog without first seeking the advice of a professional. No representation or warranty, express or implied, is made as to the accuracy or completeness of the information contained in this blog. The author and affiliated parties assume no liability for any errors or omissions.
