I'm sighted, but I decided to try the app. I've done work with OCR and usability, which will put a certain spin on my comments. Also, I'm going to write in the mode of a software tester, so I focus mostly on usability or functionality bugs, which I list in no particular order. I'll wrap with some positive comments.


I like the concept of a voice-activated app, and I know it's tricky to determine when to allow or disallow the user to interrupt the voice, but I would suggest having some means to skip past or speed up the introductory announcements. The series of introductory announcements are functionally equivalent to a series of unskippable modal dialogs. I wanted to start capturing images within 5 - 10 seconds. Thus I was already a little perturbed before I even had a chance to test functionality.

The introduction seems somewhat redundant. I'd be surprised if someone spent $2 on an app but didn't already know what the app was meant to do. So perhaps just one or two voiced lines would be sufficient before the user can start capturing images and trying some features.


Accuracy is okay, but not as good as VoiceDream Scanner, especially with uneven lighting. If you're doing some kind of preprocessing on the image, then your binarization step or thresholding step or whatever doesn't seem to adapt as well as it might to changing brightness and contrast across a page of text. As a computer science major you may have already taken one or two classes on image processing / computer vision, but if not I think it would be well worth your time. Existing image processing libraries like OpenCV are useful, but application-specific work often requires tweaking their implementations of standard algorithms. (Perhaps you're already well on your way here.)


Button functionality is sometimes hard to guess. In video / image capture mode, there is a right arrow ">" button in the lower right corner that I wouldn't have guessed meant "process the previously captured image." I was expected it to capture and process an image from the active video stream.


The "Extracting Text" notification gives no sense of how long processing may take. There's no voice announcement, so a blind user or sighted user could easily get the idea that the app has hung.


Just now, when I said "capture" and then "read," the results were from a previous image. I was attempting to reading the letters on the keys of my QWERTY keyboard.


I'm not sure what you mean by "Adding offline OCR for the first time ever." Do you mean the first time for your app? Or the first time for any app? Offline OCR, including CNN-based OCR, has existed for decades.


Although an instruction manual would be good, your app would be competing against other apps that don't require manuals. I would suggest walkthroughs instead. I haven't read a manual for any phone app in the past decade.


Out of curiosity, what is the value of having so many voices for a monthly subscription? Is that just so that someone can change up the voice every now and then? The add-on cost is a goodly fraction of the cost of a variety of streaming services (e.g. Netflix with their voiceover support). If you have marketing data suggesting $5/month is worth it then no worries, but the price seems a bit steep to me.


In the filmstrip view at bottom, when I touched an image it disappeared. I don't understand why.


The default seems to be that the text from two processed images will be combined into one text. That could be useful for text spanning multiple pages, but if someone first read a medicine bottle, and then a minute later a food label or two or three, and then other items, etc., it could be very confusing. My suggestion--and again, these are just suggestions--would be to default to a mode in which you don't save successive texts, and simply read the most recent capture. Then once the user becomes accustomed to that you could introduce stringing texts together after making multiple captures.


In general, once I got past the unskippable intro announcements and tried "capture" and "read" it wasn't clear what I should do next. Buttons on screen didn't have accompanying captions to indicate their purpose (for sighted users), though I guessed right about half the time. In short, it wasn't quite clear to me how a user would explore the app and gradually learn more.


Finally, there may be times when a blind user may not want to use voice command. (This is something I've been told in conversations about other voice command apps.) If voice command isn't used, then gestures would be a handy, silent alternative input method. I know that could add a pile of extra work, but I think you could get feedback on the desirability of gestures from your users fairly quickly. If gesture control is already built into the app, then I haven't discovered it yet, though I'd be willing to test.


"... no more than 4 buttons per screen. Crafted with VoiceOver, you simply cannot find another interface that is this simple to use." Perhaps: "you simply cannot find another OCR interface that is this simple to use." There are certainly simpler interfaces!


I think you could make further cuts to the number of buttons per screen and improve usability. Different buttons appear depending on the current operation mode--a fundamental problem of modes--and I found it difficult at times to determine what mode the app was in, and what would happen when I pressed certain buttons. (Do you have a stage diagram for the app? When video is showing and there's a previously captured image, it's not clear to me what state the app is in, which is why clicking the ">" did something unexpected.)


Again, I like the concept of voice control for an OCR app. It's harder to conduct testing nowadays when you can't sit next to one tester after the next and observe their behavior (unless you've tried that on Zoom?), so I admire your work to push out an app now.


As a user it's also good to see that you're taking into consideration the number of buttons per screen and other implementation details related to usability. I for one would be pleased with something that had less functionality but faster processing. If you identify what users do with the app 80% to 90% of the time--which perhaps you've already done--I'd recommend stripping away almost everything else.


Good luck with your app and with the fall semester!