This past weekend, AssemblyAI was a proud sponsor of HackDuke, an annual hackathon at Duke university, and HackUMBC, an annual hackathon at the University of Maryland. We saw a great turnout at the events with a combined of over 350 participants and 100 projects built in just under 36 hours.
As a sponsor, we offered our Speech-to-Text API as a resource for students to hack with and provided support for our asynchronous transcription and real-time streaming API. We also had the difficult job of choosing the best projects that used the AssemblyAI API.
Here’s a highlight of the top three projects built using the AssemblyAI API:
No need to swear! Cut out the profanity and get monetized again!
YouTube Cleaner
YouTube has advertiser-friendly content guidelines that help creators understand what videos can run ads, limited ads, and what videos will not run ads. In general, the guidelines are concerned with content and brand safety for the advertisers and viewers.
One factor that can contribute to your video being demonetized is inappropriate language, i.e., swearing and using derogatory language. As a result, many YouTube creators will try to filter themselves in videos to avoid getting demonetized, which can take hours of editing or re-shooting the video. Using YouTube Cleaner, creators can upload a video file or a YouTube URL, and the application will edit out all the profanity.
This application is built using Python, the Flask Library and the AssemblyAI API. Python and the Flask library are used to build the web server while the AssemblyAI API is used to transcribe the video. Leveraging the transcription and AssemblyAI’s Word Search feature, YouTube Cleaner automatically mutes inappropriate parts of the video. The video can then be downloaded and re-submitted on YouTube without the creator having to edit it!
2. Babel Fish
Translate any language live.
This project is inspired by the Hitchhiker's Guide to the Galaxy where a fish-looking device is used to translate any alien language.
Babel fish takes in live-stream audio using a phone connected to a Raspberry Pi and translates the audio to another language. Once the user hits the record button, the phone will begin recording in a .wav file format. Once the user hits the stop button, the audio recording is saved to the phone and sent to the Raspberry Pi to translate.
This project is built in Python using Flutter, Google Translate, and the AssemblyAI API. AssemblyAI API is used to get the transcript of the audio recorded from the phone. This transcript is then passed to the translate function using the Python library, Deep-Translator. Once the transcript is translated, it is sent back to the app.
3. AutoNote
Record and transcribe lectures to handwritten notes automatically.
This application allows users to generate handwritten notes from an audio/ video recording. It grabs audio or video from the system (e.g., a lecture recording), transcribes it with the AssemblyAI API, and generates handwritten text in a PDF output file.
It’s built in Python using the PyAudio PortAudio Library, the AssemblyAI API, and Twilio’s client library. The recording audio is performed with the PyAudio library and AssemblyAI API is used to transcribe the audio recording. Finally, the handwriting of the transcript is generated with TensorFlow. As the handwriting can take some time, Twilio’s client library is used to send the user a SMS once the handwriting is complete. Now we have a fully automated method to create written notes from lectures!
We are highly impressed with the creativity that went into all the projects, and how students used the AssemblyAI API in such creative ways! We can’t wait for another round of HackDuke and HackUMBC next year!
