Hi everyone,
I just wanted to reassure you all that the issues you are experiencing are not being ignored. The head of the development team sent the following information for me to pass on:
We've noticed an increase in reports of speech recognition "not working". First off, we know how frustrating and interruptive to your learning it can be. We apologize for the inconvenience and want to reassure you that we're working on it!
Now for some background. In the engine room, we are using the same state of the art, AI-powered speech recognition that powers all of google and apple services. However the nature of AI is that the results are ever changing and can "break" certain phrases, which is what we're seeing here. We have many ways to deal with this over the years and we're constantly patching things up to ensure the best accuracy possible. But there are still some cases where the technology struggles:
- where short words are indistinguishable from background noise, eg "ou" (French for "or")
- where the number format is mismatched, eg. twenty two dollars vs $22.
- where the phrase is not one used in everyday life, eg. sentence fragments.
- results are not absolute, they are tailored to your history of using speech services across all apps, so results can vary depending on the individual
We have ways to deal with all of these, but it's not bullet proof. Someone using our Android app in a cafe in New York may very likely get a different result from someone using their laptop in their bedroom in London. There are many factors at play.
Sorting these issues out is the next priority for our development team, which could mean changing how things work slightly. As such, in the coming weeks, we'll be in touch with everyone on this thread to see if we can isolate everyone's individual problem and find a solution for it. Please hang tight, and keep the learning in high gear!