MelodicaMate:

• 

  song notes

• 

  Home Page

• 

  song history

• 

  search page

• 

  practice page

Inspiration

Learning music alone is frustrating — especially for adults who don’t have time for classes or teachers. I wanted a tool that could listen to me, talk back, and guide me patiently, even if I’m terrible at music. The idea was to build a “teacher” that doesn’t judge, doesn’t get tired, and is always available — even for the worst music student ever.

What it does

MelodicaMate is a voice-based AI music tutor that listens to live playing or singing and gives spoken feedback. It supports both classical note names and number notation (1–7), making it easier for beginners. Users can practice freely, get real-time guidance, and receive coaching through natural voice responses.

The app focuses on learning by doing, not watching videos or reading theory.

How we built it

Flutter for the Android mobile app

On-device pitch detection to analyze live audio (no audio stored)

Flask backend hosted in the cloud

Google Gemini for generating short coaching feedback

ElevenLabs for natural, human-like voice responses

Local storage on device for sessions and limits

The system is designed to be privacy-first and lightweight.

Challenges we ran into

One challenge was finding open-source notes that fit the tone of the project and could be safely used in the app and pitch materials.

Another major challenge was that I personally cannot play music well. Because of that, the melodies and notes I played were often incorrect, which made it difficult to fully judge whether the app’s musical feedback was right or wrong. This made testing harder, but it also reflected the real experience of a beginner — which the app is built for.

Accomplishments that we're proud of

Built a working prototype that listens to live input instead of prerecorded audio

Created a conversational, voice-first learning experience

Implemented a realistic daily usage limit to support future monetization

Designed the app to remain usable even when AI services fail (graceful fallback)

Stayed honest about limitations instead of hiding them

What we learned

Real-time audio interaction is much harder than static AI demos

Simplicity matters more than perfection in early prototypes

Fallbacks and failure handling are essential for real products

You don’t need to be an expert user to build tools for beginners — sometimes being a beginner helps

What's next for MelodicaMate

Improve pitch accuracy and timing analysis

Test with real musicians and beginners for better validation

Expand beyond melodica to piano and keyboard

Add richer exercises and progress insights

Introduce an optional paid plan with unlimited coaching once the experience is stable

MelodicaMate is still early — but it’s a genuine step toward making music learning more accessible, patient, and human.

Built With

• elevenlabs
• python