Resonance
Inspiration
Music has always been one of the most powerful ways to express emotions, but creating music usually requires instruments, technical knowledge, or an understanding of music production.
We wanted to flip that idea around.
Instead of asking people to choose instruments, chords, tempo, or genres, we asked a much simpler question:
“What if you could just describe how you feel, and hear that feeling as music?”
That idea became Resonance — a generative music experience where a sentence, memory, scene, or emotion becomes an original soundtrack.
Whether someone types “I feel nostalgic thinking about late-night drives with my friends” or “A peaceful morning after a difficult week”, Resonance interprets the emotional context and transforms it into music.
What it does
Resonance converts natural language into original music.
A user simply describes a feeling, memory, scene, or mood. Resonance then:
- Understands the emotion behind the user's words.
- Extracts the emotional fingerprint, including the primary emotion, undertones, intensity, and valence.
- Transforms the emotion into a music-production brief, determining elements such as genre, tempo, instrumentation, and dynamics.
- Generates an original instrumental track based on that interpretation.
- Delivers the generated music for the user to listen to and keep.
The entire experience is designed around one principle:
No instruments. No music theory. Just sentences.
How we built it
We built Resonance as an AI agent pipeline that connects language understanding with generative music.
Emotion Intelligence
The first stage analyzes the user's text and identifies the emotional characteristics hidden inside it — not just keywords, but the overall emotional context, intensity, and tone.
Prompt Alchemy
The emotional information is then transformed into a structured music-production brief.
The system determines appropriate characteristics such as:
- Genre
- Tempo
- Instrumentation
- Mood
- Dynamics
- Musical atmosphere
This creates a bridge between human emotion and musical instructions.
Lyria Sound Engine
The final stage sends the generated musical direction to Google's Lyria music generation model, producing an original instrumental track.
The application was built using Google ADK, Gemini, and Lyria, with the frontend providing a simple interface where users can describe what they want to hear.
Challenges we ran into
One of our biggest challenges was translating something as subjective as human emotion into something as structured as music.
A sentence can contain multiple emotions at once. For example, nostalgia can be both happy and sad, while a sentence describing loneliness might still contain a feeling of hope.
We therefore had to think beyond simple sentiment analysis and create an emotional interpretation that could influence multiple musical dimensions.
Another challenge was creating a pipeline where the output of one AI stage could meaningfully influence the next stage rather than simply chaining independent AI calls.
We also had to balance:
- Emotional accuracy
- Musical quality
- Generation time
- Simplicity of the user experience
- Consistency between the user's description and the generated track
The goal was not simply to generate “a song,” but to make the music feel connected to the words that created it.
Accomplishments that we're proud of
We are most proud of turning a very simple interaction — typing a sentence — into a complete creative experience.
The user doesn't need to know:
- Music theory
- Chord progressions
- Instrumentation
- Audio production
- Prompt engineering
They simply express themselves.
We also created a multi-stage AI pipeline where language understanding, emotional reasoning, prompt generation, and music generation work together as one experience.
Most importantly, we created something that feels more like communicating with music than using a traditional music-generation tool.
What we learned
We learned that generative AI becomes much more interesting when multiple capabilities are combined into a meaningful workflow rather than used independently.
We also learned that prompt engineering is not only about writing better prompts. It can be about building systems that transform information from one domain into another.
In Resonance, we effectively created a translation layer:
Human language → Emotion → Musical intent → Music
We also learned that emotional context is highly nuanced. Two sentences can describe the same emotion while requiring completely different musical treatments depending on their context.
What's next for Resonance
We see Resonance becoming more than a text-to-music generator.
Some of the directions we want to explore next include:
🎙️ Voice-to-Music
Allow users to speak instead of type and transform their voice and words into music.
🎥 Video-to-Soundtrack
Analyze a video and automatically create a soundtrack that matches its emotional progression.
🧠 Adaptive Music
Generate music that evolves as the user's emotional state or story changes.
🎮 Interactive Experiences
Use Resonance to create dynamic soundtracks for games, interactive stories, and virtual experiences.
🎨 Multimodal Emotion
Combine text, images, video, and voice to understand a richer emotional context before generating music.
👥 Collaborative Creation
Allow multiple people to contribute their emotions or stories and create a shared soundtrack.
Our long-term vision is simple:
Resonance should make music creation as natural as expressing a feeling.
Describe it. Feel it. Hear it.
Built With
- adk
- api
- cloudrun
- gemini-3.5
- google-cloud-console
- lyria
- vertex
Log in or sign up for Devpost to join the conversation.