An hour ago, I was sitting at a banquet table with some fellow Cobalters, listening as presenters at the Project Voice Awards presented awards to a variety of accomplished people and companies in our industry. There were awards for best automotive application, best Alexa skill, best journalist covering voice, and others. Nearing the end of the presentations, the final award to an individual was to be given to the Voice Pioneer of the Year. I was among 19 nominees, many of whom had done impressive work this year, and I had approximately zero expectation of winning. Then came the announcement: “And the winner is… Jeff Adams of Cobalt Speech!” I was a surprised and unprepared winner, and as I accepted the award, I mumbled a short, impromptu acceptance speech about the award really belonging to the whole team at Cobalt.
After I sat back down, I began wondering: Was I really a pioneer? What does that even mean?
The real truth is that I’m not the pioneer. I’m fortunate to be surrounded by an absolutely amazing team at Cobalt that really do deserve this honor, and I share it with them. It is they who have explored new technologies in speech, new approaches to working with partners, and new applications, while I have cheered them on.
Here, at the beginning of 2020, it is perhaps appropriate for me to share the spotlight with the team, and to look back at how Cobalt has been a real pioneer in 2019. Here are just a few highlights of the past year:
- We have made great improvements to the ease of deployment and integration of our Cubic ASR engine, for both on-prem/server and embedded applications. We now have tools to help developers customize lexicons and language models to give them even more control and independence.
- As described in a previous blog post, we explored how to get unprecedented ASR accuracy by adapting language models to specific domains. We have done similar work to this throughout the year with various customers (under NDA, so no details here), adapting language models, acoustic models, lexicons, grammars, NLU models, and more.
- We added several new languages to our ASR arsenal in the past year, including German, Russian, Korean, and improved models for our existing languages.
- We released our Telefol engine for both off-line and real-time indexing and keyword spotting of speech.
- We found novel ways to partner with customers using equity or profit-sharing (for small, cash-starved startups).
- We developed the most accurate diarizer on the commercial market. (A diarizer detects when speakers change in a conversation. It is often used to convert single-channel phone call recordings into the two separate speaker tracks.)
Truth be told, the most interesting and innovative things we have done this year will remain unknown because they were done under NDA with our trusted and fabulous customers. I really wish we could tell the world about some of those projects, but alas, we can’t.
In none of these projects, however, was I the pioneer. If I deserve any credit, it is for establishing a company setting where the best minds in the industry can find a home and do pioneering work on the challenging projects our customers bring to us. But let’s be real — the true pioneers are the ones who do the work. To my pioneering colleagues at Cobalt Speech & Language: Thank you. I respect you; I honor you; I am honored to work with you.
Here’s to even more pioneering in 2020 and beyond!
About the Author
Jeff Adams is the founder of Cobalt Speech & Language. In his former professional life, he led research teams at Nuiance, Yap, and Amazon. He is the author of about 25 patents and several published research papers.
He’s also a very nice guy.