As was mentioned in the last update, Shane and Zack have been working together to integrate the software and create a fully working software demo. I’m happy to report that they’ve found success. The software is now fully integrated and the entire pipeline is now running from start to finish. Currently this process is hard coded to run and process 50 image scans at a time and then stop, so the next steps with this is to write software that will allow this process to run indefinitely.
Going through this process we’ve learned that we are in fact able to do the OCR and TTS faster than the audio can play, confirming our earlier tests about reading speed. It is currently taking around 0.3 - 0.5 seconds to complete the OCR/TTS pipeline, giving us a reading speed between 120-180 words per minute. To give you an idea how that tempo will sound, here is a YouTube video of someone reading at different speeds. (Start at the beginning for 120 wpm, start at 14:25 for 180 wpm)
Here is a video of our software running and the output we’re getting. For reference, the passage that is being scanned is “Mary Had a Little Lamb.” Please keep in mind that the OCR model has yet to be trained, which is why it is not accurately picking up all the words. However, based on prior tests we've conducted, we are confident that we’ll be able to achieve accuracy >98% once the OCR is fully trained.
We’ll be training the OCR using a machine learning model. We’ll be able to scan a book and match the characters/words up with it’s eBook counterpart. The way this will work is the model will get the data input from the camera, guess what word it should be, check it against the eBook for reference, and use that as feedback to improve itself.
Those are the main things that Shane and Zack are working on at the moment. There are still some small things that will also need to be done such as programming the button functionality and the LED sequencing, but that should be easy. They’re expecting to have a fully functional software demo for you within the next few weeks.
We’ve been working to make improvements to the 3D model of our plastic enclosure recently and we received a new 3D print about a week and a half ago.
There were both some good and bad things about this new model, but it taught us quite a bit and will help us make improvements for future iterations.
The ribs added a lot of strength as hoped, the button size, shape, and location are all great, the battery partitions were perfect, the shape of the lens mount was good. The lip groove extension worked well and it fit together a lot better.
The lens mount was too thin and weak, still having trouble routing the cables and making it fit, The button fit was not good. It all fit together but there was no range of motion. Assembly was difficult.
We’ve already made improvements to the design and sent off for another 3D print. We should receive it around the middle of next week. In this updated design we made it easier to assemble by putting all the components on one side. We strengthened the lens mount and made it to where there would be no undercut to allow injection molding. We made it 4 mm longer to give ourselves more room to route the cables and made changes to the button to give it a larger range of motion.
As we’re making improvements we’ve been running the model through Protolab’s automated quote generator to estimate costs and see what changes will need to be made prior to tooling up. As you can see below, there are a few required changes, and we’re working those changes into each new design iteration.