jc-cr / FindMe

A package for determining the relative direction of a voice and orienting a camera towards that voice.

FindMe

... a gaze-mimicking package for robots.

To Do

• Setup BBB
  • Audio input
  • Tensorflow lite
• Test keyword spotting examples
• Record Audio samples
  • Record a few audio samples from 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315° at varying distances
• Combine STFT audio processing and fuzzy logic azimuth estimation
  • STFT: See if we can identify frequecny spikes when speech is from a certain direction
  • Fuzzy Logic: Once we identify some characteristics, we will use fuzzy logic to estimate direction of audio
• Combine keyword spotting and azimuth estimation
• Control Servo with azimuth results
• Create demo with realsense and base

Overview

This package utilizes a single microphone and a BeagleBone Black board to detect the relative location of human speech and automatically orient a Realsense camera towards the sound source. When a human voice is detected, the software calculates the direction of the sound and adjusts the camera's orientation using a servo.

Hardware

• Digital Servo
• Common microphone
• Good microcontroller or SBC
• A USB camera

Installation

1. WIP

Usage

1. WIP

Examples

1. WIP

Acknowledgment

See Work Referenced. Workflow improved by ChatGPT.

License

This project is released under the MIT License. See the LICENSE file for more information.