# SoundScope

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Keith Rwezahura](https://cerebralvalley.ai/u/Kamahura)
- **GitHub:** https://github.com/rwezzy/soundscope/tree/soundscope-v3
- **Demo video:** https://youtu.be/chX38vXhlto
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5359

SoundScope is a sound-awareness tool that shows you where a sound came from, what it was, and who said what.

For people who are Deaf or hard of hearing, or anyone in a loud space, like emergency operators at a rescue site, or birdwatchers in an ambient forest, sound can lose its precise direction as we ordinarily perceive it.

SoundScope gives this direction back by sharing; or transferring, the burden of auditory perception to your eyes. A Seeed Studio reSpeaker XVF3800 (4 mics) hears each sound a fraction of a millisecond apart; physics (SRP-PHAT with echo handling) turns those delays into a direction, and an Orbbec Gemini 2 depth camera marks the exact spot in 3D on a live heat map. No image detection decides where a sound came from.

Meta models add the meaning:
- Muse Spark 1.3 names the sound from its spectrogram and a camera crop of the spot, and answers plain-language searches like "when did my phone ring?", jumping back to the saved scene and audio.
- Muse Voice Transcribe turns speech into color-coded subtitles at the bottom of the screen, with an arrow showing who is talking, and alerts you when someone says your name.
- SAM 3.1 outlines the object at the sound's location.

When it isn't sure, it doesn't raise a false alarm: it shows a calm "Possible doorbell?" card, and your Yes/No answer teaches it that sound for next time. You can name any sound or person and it remembers. Claps are recognized locally for free, and a budget cap keeps model spending low.

Uses: Deaf and hard-of-hearing users, search and rescue (which way to dig), birdwatching, and finding a lost phone.

Built with Python, OpenCV, NumPy, and the Meta APIs. It is a sound-awareness aid, not a medical device.

## More from Meta Global AI Developer Hackathon

- [AgroSentinel Field](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5356)
- [Remtekindo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5357)
- [MetaMate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5358)
- [VoiceGuardAI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5360)
- [Kody Anh Nguyen](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5361)
- [Wyrd](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5362)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5359. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt