Introducing 

Prezi AI.

Your new presentation assistant.

Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.

Loading…
Transcript

Sound Recognition

and Localization

ARA Sound Recognition

SUMMARY

ABOUT

Our Problem

Introduction

Demo

OUR GOAL

Help deaf and hearing loss people.

Impact

TO Delete

KEY RESULTS

Measuring success

Measuring success

Key components

Key components

METHODOLOGY

HYPOTHESIS

DIVING DEEPER

LITERATURE REVIEW

What is our approach

Added Value

ADDED ENGINEERING VALUE

Ema'a organization

APPLICATIONS

Baby Phone

What World

DO

MyEarDroid

Shazam

Our proposed Ideas

Theoretical ideas proposed

METHODOLOGY

HYPOTHESIS

Methods : Finger Print

Methods : Classification

Original Audio Signal

FingerPrint

Extraction

PrePare Data

collect data

Extracted

Fingerprints

Training

model

FingerPrint

Extraction

FingerPrint

Matching

Fingerprints

+

Metadata

Captured

Signal

Model

Selection

Evaluation

Matched

Metadata

Methods : Classification

PrePare Data

collect data

Training

model

Model

Selection

Evaluation

Methods : Finger Print

Original Audio Signal

FingerPrint

Extraction

Extracted

Fingerprints

FingerPrint

Extraction

FingerPrint

Matching

Fingerprints

+

Metadata

Captured

Signal

Matched

Metadata

FFT

N envelope time

domain time segement

Covariance Matrix

Coherence

Number of sound Resource

Source Detection

Distance From Audio source

Calucate Time

Difference

Sound Source Direction

Azimuth

Angle

ATD Algorithm

DIVING DEEPER

HYPOTHESIS

Methods : Finger Print

Methods : Classification

Original Audio Signal

FingerPrint

Extraction

PrePare Data

collect data

Extracted

Fingerprints

Training

model

FingerPrint

Extraction

FingerPrint

Matching

Fingerprints

+

Metadata

Captured

Signal

Model

Selection

Evaluation

Matched

Metadata

FINDING

OUR SYSTEM

Fingerprints

Features

Feature

Extraction

Model

Match/

Not Match

sound

type result

Sound

Direction

Implementation

• Languages : Java + Python + PHP

• Libraries: Spicy + Librosa + Numpy + Scikit-learn

• Frameworks: Laravel + Android Studio + Jupyter

• Using : Git-hub + Heroku online server

Classification Steps

1st step : Collecting Data

Urban Data Sound

8732 audio sound with ".wav" extension

We added New audio sound to urban dataset

Divided All sounds into 10 folds

To get Our Final Classes

Final Classes

Air-conditioner

jackhammer

Baby Cry

Car-horn

siren

Door Squeak

Children-playing

Street-music

Water Tape

Dog-bark

Glass-Broken

Footsteps

Engine-idling

Tones

Knock Knock

Gun-shot

Scream

People Talk

2nd Step: Processing Data set

A. Signal Pre-processing

like : MFCC

B. Signal Projection

like : Fourier Transform

C. Feature Extraction

like : Mel-frequency cepstral coefficients(MFCC)

Chroma, melspectogram, spectral contrast, Tonnetz

D. Feature Optimization

like : Normalization

3rd Step: Training Model

Algorithms we used to create our training model :

1- Multi-Layer Perceptron (MLP)

2- Linguistic Regression (LR)

3- K Nearest Neighbor (KNN)

4- Decision Tree (DT)

5- Support Vector Machine (SVM)

6- Random Forest

3rd Step: Training Model

4th Step: Model Selection

RECOMMENDATION

FUTURE LOOK

Use Microsoft Hololens With Augmented Reality

Adding Speech To Text

Use bracelets for hands on a slight shake to guide the user to the sound Direction

TEST

THE END !!

Learn more about creating dynamic, engaging presentations with Prezi