Hookd

Inspiration

Short-form content is the new way for musicians and businesses to break into several industries.

Finding the perfect hook is a £100-million problem for artists, labels and brands, with countless hours lost to manual review.

Hookd is an AI-powered crystal ball for virality, helping creators, labels and businesses instantly identify the moments that make music viral or the perfect soundtrack for their next campaign.

What it does

Hookd predicts which 10 seconds of an uploaded song are most likely to go viral, helping artists and advertisers capture attention before anyone else does.

How we built it

We built Hookd using Python, leveraging Librosa to extract key audio features (such as tempo, energy, spectral contrast and harmonicity) from each song. We defined a set of metrics that capture what makes a clip “viral,” including rhythmic intensity, emotional variance and memorability.

These features were applied across a large dataset of Deezer tracks. We then trained a Random Forest Regressor to predict a “virality score” for optimised 10-second windows.

Finally, our model identifies and visualizes the segment of a song most likely to go viral, providing creators and brands with an instant highlight reel of their track’s most powerful moments.

For the application interface, we built a web front end using HTML and CSS, creating a clean and intuitive dashboard for users to upload tracks and view viral segment predictions. The back end is powered by Flask, which handles audio processing, feature extraction and model inference, allowing seamless interaction between the user interface and our machine learning pipeline.

Challenges we ran into

None of us had prior experience with Librosa or large-scale audio analysis, so extracting meaningful features from raw sound took significant trial and error and reading. We also faced issues integrating API's for our training data, running into many issues when initially using Spotify's API before switching to Deezer's.

Another major challenge was understanding music industry terminology and conventions — figuring out how labels and creators musically define a “hook” or a “viral moment” required us to translate technical signals into creative language that makes sense to artists and brands.