WORK ACCOMPLISHED:
This week, a lot of time was spent testing the capabilities of different models for Google Speech-to-Text AI for extracting the destination for a journey from the user’s voice commands. After testing the different models, the decision was made to use the Chirp 2 model with model adaptation. The use of model adaptation is very important as it improves the accuracy of the recognition system. When testing, it was noticed that the system struggles with words that sound very similar such as “weather” and “whether”. As a result, with model adaptation, I can set a “boost value” for a phrase such as “weather” so that the system is optimized for identifying specific phrases.
Additionally, the logic for navigation suggestions was developed a bit and we have worked on some code that uses R-tree algorithms for identifying the appropriate navigation instruction based on the user’s real-time GPS location.
Snippet of Navigation code using R-tree algorithm:
Sample output:
PROGRESS:
I am currently on progress with my work.
NEXT WEEK DELIVERABLES:
For next week, we will work on building out the navigation system and handling the case when the user is completely off path. Also, we will be working on fabrication and 3D printing for the bike mount. Also, we will start looking at how to convert text of navigation to audio.