In the age of digital memories, SundAI Club has developed a unique solution to capture those fleeting moments of hilarity that often slip away after long conversations or game nights. Visit Chuckle Box, our latest AI hack that turns hours of audio into a curated list of the funniest moments.

Project Overview:

“Chuckle Box” aims to solve a common problem: remembering and sharing the best jokes and witty exchanges from lengthy social gatherings. Our system processes up to four hours of audio, identifying laughter and extracting the surrounding context to create a highlight reel of humor.

Here's how it works:

  1. Recording: Users capture their conversation using a simple iPhone recorder.
  2. Uploading: The audio file (mp4 or mp3) is uploaded to our application.
  3. Transcription: Our system transcribes the entire conversation.
  4. Laughter Detection: A machine learning model identifies moments of laughter in the audio.
  5. Highlight Extraction: Large Language Models (LLMs) analyze the transcribed text around laughter moments to identify and summarize the jokes or punchlines.
  6. Presentation: Users receive a list of summarized funny moments, ready to be remembered and shared.

Technical Breakdown:

Front End Team: Our UI experts focused on creating an intuitive interface for audio upload and result presentation. The challenge was to make the process as seamless as possible, ensuring users could easily access their humor highlights.

Back End Team: This team split into two sub-groups:

  1. Machine Learning Sub-Team: They tackled the core of our project - accurate laughter detection. Using an open-source machine learning model and a GPU instance for faster inference, they fine-tuned the system to recognize various types of laughter in different audio environments.
  2. Integration Sub-Team: This group managed the transcription process and ensured smooth end-to-end integration. They worked on optimizing the pipeline to keep runtime within acceptable limits, a significant challenge given the large audio files and complex processing involved.

The most innovative aspect of “Chuckle Box” is its combination of laughter detection and context understanding. By using LLMs to analyze the conversation around detected laughter, we're able to provide meaningful and genuinely funny highlights, not just moments of noise.