TxtScribe

• 

  Input image

• 

  Traditional handwriting to text output

• 

  TxtScribe output. Using Cohere's NLP APIs, we are able to sanitize the output of the the ml model, hence resulting in a far better output.

Inspiration

I was inspired by my experiences during my first few weeks of university. By the end of these two weeks, my phone's gallery was full of pictures of class notes, posters, stories and other interesting stuff I had found across the campus. I wanted a quick way to convert all those pictures into text files so that I can easily save and search through them when needed. I tried using online handwriting to text converters, but every single one of them failed at creating a coherent transcription. There were always missed words, some random errors, weird spacing, or they outright couldn't understand my handwriting. When I learnt about cohere's NLP APIs, I realized that it could fill in that gap, and provide me much better transcripts than what I could otherwise generate.

What it does

TxtScribe allows you to click or upload a picture, and it will automatically generate a transcript for the text. Other than that, it'll also provide you with a quick summary of the passage, allowing you to read over the entire paragraph easily.

How we built it

TxtScribe was built by me, over the weekend, with Python, Flask, and HTML5. I leveraged the power of google cloud to run the handwriting detection algorithm, which is then drastically enhanced with the power of cohere's APIs. I also use cohere's NLP Generation API to generate a quick summary of the text.

Challenges we ran into

One major challenge that I faced while developing TxtScribe was integrating all the APIs into a common backend, but thankfully cohere's impressive support for a wide variety of languages and development stacks meant that I was able to easily integrate it into my backend code, which was written in Python.

Built With

• cohere
• flask
• google-cloud
• html5
• python