Lyra | Devpost

Inspiration

Lyra was born out of our frustration with trying to find songs stuck on the tip of our tongue. While apps like Shazam work great when music is playing, they can't help when you only remember a few lyrics or a melody. We wanted to build something that lets you find any song just by singing or saying what you remember, and then use that as a foundation to build your music library. Just like how a single lyric can lead you to a full song, we believe one song can be the starting point for an entire personalized playlist.

What it does

Lyra is a simple yet powerful tool that lets you:

Record yourself singing or speaking lyrics you remember See multiple potential song matches with artist information Listen to the songs through Spotify integration Create playlists based on your discovered songs

The process is intuitive: tap the microphone, sing or speak lyrics, and let Lyra do the rest.

How we built it

We built Lyra using a combination of modern technologies:

Backend: Python Flask for handling audio processing and API integrations Frontend: React for a responsive and dynamic user interface Voice Recognition: Deepgram API to accurately transcribe speech to text Lyrics Matching: Genius API to search their database and find matching songs Music Playback: Spotify API to play previews and generate playlists

The core workflow is straightforward – we capture audio from the user's microphone, send it to the Deepgram API for transcription, then query the Genius database to find the best song matches. Finally, we display the results and enable playback through Spotify.

Challenges we ran into

Building Lyra wasn't without its hurdles:

Working with audio quality variations from different microphones Matching partial or misremembered lyrics to the correct songs Filtering out remixes and covers to prioritize original tracks Learning to work with multiple unfamiliar APIs Optimizing the backend to handle audio processing efficiently

Each challenge pushed us to better understand audio processing and natural language matching techniques.

Accomplishments that we're proud of

We're particularly proud of creating a seamless experience from voice input to song results. The ability to identify songs from just a few seconds of sung lyrics, even when they're not perfect, feels almost magical. Our UI design makes the whole process intuitive, and the integration with Spotify makes it immediately useful.

What we learned

Audio processing in web applications Working with speech-to-text technologies Efficiently querying music APIs Building responsive user interfaces for voice interactions