AURAView source

EXPERIMENTAL · LOCAL DEVICE PROTOTYPE

A voice that can
look when asked.

AURA explores hands-free interaction with spoken questions and an optional camera request, using a Python command-line prototype.

Explore the workflow Browse the code

Presentation site only. It does not access a microphone, camera, or AI provider. Run the prototype on a configured local machine.

EXISTING PROJECT DEMOAURA / 01
Existing AURA project demo animation
Existing demo asset from the repository. It is illustrative; the live device workflow was not re-run for this presentation.
WHAT THIS REPO EXPLORES

Speech input → intent routing → optional visual context → spoken response

One request per wakeword

01 / INTERACTION

One voice turn,
from wakeword to reply.

The application runs in a terminal and coordinates local device access with external speech and AI services.

01
◉

Wake

Porcupine listens locally for the configured keyword model.

LOCAL DEVICE
02
≈

Speak

Microphone audio is streamed to Deepgram for transcription.

EXTERNAL SERVICE
03
⌁

Understand

Groq routes the request; an explicit visual request may invoke the camera path.

EXTERNAL SERVICE
04
♫

Respond

A response is spoken through the local audio output.

LOCAL DEVICE

02 / SYSTEM MAP

Local devices.
Remote inference.

The camera route captures a still image only for a visual request and sends it to Groq without an in-app preview. Review provider privacy terms before use.

Read architecture in the README

03 / GET STARTED

Bring your own
devices and keys.

This is a hardware and credential-dependent prototype. The complete flow was not verified during this review.

  1. 1

    Use Python 3.10 or newer and install the repository requirements on your target platform.

  2. 2

    Configure Groq, Deepgram, and Picovoice credentials in config/.env.

  3. 3

    Obtain the platform-specific Porcupine keyword file; it is not included in the repository.

  4. 4

    Connect a microphone, camera, and speaker, then run the CLI.

LOCAL TERMINAL
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
cp config/.env.example config/.env
python main.py

Audio packages and devices may need platform-specific setup.

BOUNDARIES

A prototype, with
known constraints.

  • Not evaluated for accessibility. It is not a validated assistive tool.
  • Not for safety-critical use. Model output can be wrong, including about labels and medicines.
  • External providers receive data. Speech goes to Deepgram; transcripts and optional images go to Groq.
  • Not hosted as an app. Vercel serves this static project presentation only; device features remain local.