We were moved by the need to connect the sponsor with their sponsored little ones in need all over the world.

What it does

It allows the sponsor to talk, in his/her native language, to their sponsored kid also in his/her native language. It also censors unwanted words or images and reports repeated offenders.

How we built it

We built it using AWS Transcribe to read the audio and turn it into text, we then used AWS Translate to translate the text into english, if not already in english. We use comprehend to look for the sentiment and if it's negative, we replace it with a custom message. If the message is not negative, we use AWS Polly to translate the text from english to the childrens language.

Challenges we ran into

AWS Transcribe was very slow. We found out that a faster near real-time version exists, but in Java, and we chose to make a browser based platform in Javascript; we decided not to integrate it yet.

AWS Comprehend wanted us to provide it with 1000 documents of profanities. We didn't have them, so we compromised on doing something simpler.

Accomplishments that we're proud of

What we learned

AWS Transcribe AWS Translate AWS Polly AWS Comprehend

What's next for Table21

We wanted to use the AWS SDK to use AWS Transcribe Streaming, which is faster than AWS Transcribe and would allow us for real time voice to text. Unfortunately, the API is only available for Java and we are running a browser application. We hope that AWS releases the SDK for javascript so that we can leverage it and improve our speed.

We think we can improve the usage of Amazon Comprehend to identify profanity, but we need to gather samples of profanity in the english language, and we didn't have to a document of profanities. We decided to use the sentiment instead, as a placeholder for the profanity analyzer.

Code links

Built With

Share this project: