Google is betting that “Open” is the future of artificial intelligence.
The Mycroft A.I. team was very excited to see Google publish their machine learning libraries as a new open source initiative. Dubbed “Tensor Flow” this tool set includes both Python and C++ libraries for creating machine learning datasets, training models and interpreting data based on established models.
According to the company Tensor Flow is used heavily in Google’s speech recognition platform as well as their image search system. It is also an integral component of their real time translation software stack.
Google’s publication is great news for Mycroft. Over the past several months we have been evaluating a number of solutions for STT including Sirius, wit.ai and Google’s translation services. Though several of these systems work well, we’ve been a bit leery of them because we don’t want to send Mycroft user data to a third party. Ideally we want be run our speech to text translation software locally or ( if the processing requirements prove to be too high) in our own data center.
Based on our evaluation so far Google translate has consistently been faster and more accurate than the others. That is why we are so excited to see the underlying technology released into the open source community.
Over the next couple days our A.I. expert Jonathan D’Orleans will be evaluating Tensor Flow to see if it meets our needs here at Mycroft. If it does, we are going to begin the process of training in-house models with a goal of running the speech to text system within our data center. Ideally this will improve user privacy, reduce latency and improve accuracy.
CTO of the Mycroft project. A long time open source software developer Ryan has contributed to the Solus Project, GtK, DuckDuckGo and many others. He is a former systems administrator for the Northeast Kansas Library System. Ryan has extensive experience with datacenter management including the construction and maintenance of high availability virtualization systems.