Bumblebee Smart Speaker

Inspiration

Many children struggle with speech challenges like rhotacism, where “R” sounds are replaced with “W” or other sounds (e.g., “rabbit” → “wabbit”).
Traditional therapy can feel repetitive or intimidating. We wanted to build a fun, AI-powered smart speaker that makes practicing speech engaging, playful, and supportive — a tool kids actually want to use.

What it does

Bumblebee works like a kid-friendly Alexa. It encourages children to:

Tell stories, answer questions, and play word games.
Practice targeted speech exercises for “R” sounds and other difficult phonemes.
Receive instant, positive feedback in a gamified format — turning practice into a fun challenge.

Behind the scenes, Bumblebee transcribes speech into a phonetic alphabet using Hugging Face models and custom ML pipelines. This lets us track improvement over time and measure clarity of “R” sounds, while also keeping interactions lighthearted and encouraging.

How we built it

Gemini & Google Vertex AI → speech analysis, natural conversation, and personalized feedback.
Firestore → stores user progress, achievements, and exercise history.
Hugging Face models → phonetic transcription + rhotic sound detection.
Custom voice interface → safe, approachable, and designed for children.

Challenges we ran into

Detecting subtle pronunciation differences in noisy environments.
Keeping feedback positive and motivating without discouragement.
Designing child-safe prompts that balance fun storytelling with structured practice.

Accomplishments that we're proud of

Built a working pipeline: speech input → phonetic analysis → feedback loop.
Created exercises that kids genuinely enjoy, blending therapy and play.
Designed a system parents, teachers, and therapists can trust.

What we learned

Speech therapy benefits from gamification and positive reinforcement.
Tracking small improvements over time builds confidence.
AI can complement (not replace) therapists by making practice more accessible at home.