SlideShare a Scribd company logo

Kinect Microphone Array case study

Muhammad Imran, profile picture
Muhammad Imran

The document outlines a detailed procedure for measuring and analyzing localization performance using a Kinect microphone array. It covers data acquisition methods, array performance analysis, and experimental evaluations with various sound sources, focusing on accuracy and robustness of localization. Various statistical metrics are used to quantify performance based on ground truth angles and the results demonstrate high accuracy across different angles.

Engineering
1 of 57
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
PROCEDURE FOR LOCALIZATION AND ARRAYS DESIGNING KINECT MICROPHONE ARRAY CASE STUDY FOR LOCALIZATION PERFORMANCE 2016-10-08 MUHAMMAD IMRAN
CONTENTS • Measurement Procedures (Data Acquisition) • Array Performance Analysis • Localization Performance • Kinect Case study for Localization Performance
PROCEDURE Measurement Procedures (Data Acquisition)
MEASUREMENT PROCEDURES (DATA ACQUISITION) 1. Measurement Setup a. Data Acquisition HWD b. Source Selection • Wideband Source • Narrowband Source (Single Frequency) • Speech Source c. Source Directivity • Narrow (±20) • Speech Source d. Geometry Plan • Placement of source and receiver • Alignment of Array
DATA ACQUISITION HARDWARE a. Genelec 8030 • 20 seconds • Sampling = 44.1 KHz (Prefer) • 860” Frames (if Frame size is 1024) • 860” DOAs for One position b. Kinect Array • 04-CH: Linear Array • Non-Uniform • Windows Direct Sound • Interface • Source: Professor Edgar Choueiri, 3D3A Lab, Princeton University • For Details: Microsoft Corporation
PLACEMENT AND ALIGNMENT • Making Grid (Physical) • Thread Network • Protractor • Separation between elements • Error within 1 mm • Axial Placement • Alignment Errors • Within the limits of (Hz, Vt) • -0.5o to +0.5o • Environmental Factors • Temperature (STP) • Humidity (STP) • Back Ground Noise Level <= 20dB (±5dB)1
PROCEDURE FOR DATA ACQUISITION • Kinect alignment to Loudspeaker • With Magnetic Compass • Exciting Sound Sources: 1. Pink Noise (Wideband Source) • Duration = 10 sec • Measurement Angles = -90 to + 90 (Step: 05o) • For Linear Array 2. Sine 500, 1000, 1500 Hz • Duration = 10 sec • Radom angles (05-10 angles) 3. Speech (Anechoic) • Duration = 10 sec • Measurement Angles = -90 to + 90 (10o)
ARRAY DIRECTIVITY • Figure 1: (Ref: 02): Microphone array directivity pattern in three dimensions • What to measure (Ref: 01) (Formulae and Matlab Example Code) 1. Microphone Sensitivity 2. Microphone Noise and Output SNR 3. Array Directivity Pattern 4. Array Frequency Response 5. Array Directivity Index Figure 1 Reference: 1. Chapter: 03: Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009 2. MIT Media Lab: Kinect Array Directivity
ARRAY DIRECTIVITY • Figure 2: A directivity pattern of Kinect (Azimuth @ 0o Elev.) Figure 2a Figure 2b Reference: 1. Chapter: 03, Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009
DIRECTIVITY INDEX COMPARISON • Figure 3: A directivity Index of different number of microphone • More Microphones more directivity • More computational cost • Less Microphones Less directivity • Less computational cost Directivity Index 0 2 4 6 8 10 12 Uni-directional Tw o element Four element Microphone/Mic Array DI,dB Figure 3 Reference: 1. Chapter: 05, Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009
PROCEDURE Localization Performance
PERFORMANCE ANALYSIS PROCEDURES (DOA) 1. Flow of computation a. Data segmentation b. Frame by Frame Computation a. Localization Algo (GCC-PHAT) b. Individual Pair Computing c. Combining Pairs 2. Verification Parameters a. Localization Accuracy b. Localization Resolution c. Localization Stability (Standard deviation) (The standard deviation tells the dispersion of data) d. Robustness (Root Mean Squared Error: RMS) 3. Plotting and Describing Results a. Histograms, Polar Plots
FLOW OF COMPUTATION Flow of computation a. Data segmentation b. Frame by Frame Computation a. GCC-PHAT b. Individual Pair Computing c. Combining Pairs c. Recorded signal Length a. 10 sec d. Sampling a. 44100 Samples/Sec e. Total Frames a. 430 (@ 1024) f. Total DOA a. 430 g. Recommended minimum = 200 CH1 CH2 CH3 CH4 Signal Length (10 sec)
VERIFICATION PARAMETERS 1. Plotting and Describing Results a. Histograms, Polar Plots 2. Verification Parameters 1. Localization Resolution 2. *Localization Accuracy (Error) 3. *Localization Stability (Standard deviation) 4. **Confidence Level Reference: *Chapter: 06, Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009 **https://en.wikipedia.org/wiki/Confidence_interval
Kinect Localization Performance Experimental Evaluation
MEASUREMENTS • Source: Pink Noise • Ground fact angle: -90o to + 90o • Measured duration: 10 to 20 sec 1. Pink Source 2. Speech 3. Anechoic Speech Statistics • Mean • STD • Error w.r.t. Ground angles • Range (Target Range = ±5 @ Kinect) • Stability / Robustness • In Range: % • Out Range: %
• Source: Pink Noise • Ground fact angle: 0o • Measured angles: (533) 1. Mean: -0.12o 2. STD: 0.11o 3. Error: 0.13o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.6o a. In Range: 100% (533) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 5o • Measured angles: (533) 1. Mean: 4.97o 2. STD: 0.18o 3. Error: 0.028o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1o a. In Range: 100% (533) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 10o • Measured angles: (543) 1. Mean: 9.87o 2. STD: 0.134o 3. Error: 0.13o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.6o a. In Range: 100% (543) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 15o • Measured angles: (133) 1. Mean: 15o 2. STD: 0.25o 3. Error: 0.004o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.3o a. In Range: 100% (133) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 20o • Measured angles: (533) 1. Mean: 20.16o 2. STD: 0.19o 3. Error: 0.15o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.9o a. In Range: 100% (533) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 25o • Measured angles: (541) 1. Mean: 25o 2. STD: 0.25o 3. Error: 0.001o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.4o a. In Range: 100% (541) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 30o • Measured angles: (534) 1. Mean: 29.7o 2. STD: 0.17o 3. Error: 0.29o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.8o a. In Range: 100% (534) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 35o • Measured angles: (519) 1. Mean: 34.05o 2. STD: 0.17o 3. Error: 0.95o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1o a. In Range: 100% (526) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 40o • Measured angles: (526) 1. Mean: 39.08o 2. STD: 0.19o 3. Error: 0.91o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1o a. In Range: 100% (526) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 45o • Measured angles: (190) 1. Mean: 43.9o 2. STD: 0.37o 3. Error: 2.9 4. Robustness: 100% (Target Range = ±5) 5. Range: 3o a. In Range: 100% (190) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 50o • Measured angles: (215) 1. Mean: 48.02o 2. STD: 0.5o 3. Error: 2o 4. Robustness: 100% (Target Range = ±5) 5. Range: 3o a. In Range: 100% (215) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: 55o • Measured angles: (195) 1. Mean: 54.1o 2. STD: 1.5o 3. Error: 0.8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 7o a. In Range: 94% (204) b. Out Range: 6% (11) Results
• Source: Pink Noise • Ground fact angle: 60o • Measured angles: (200) 1. Mean: 58.2o 2. STD: 2.34o 3. Error: 1.7o 4. Robustness: 100% (Target Range = ±5) 5. Range: 8.9o a. In Range: 87% (176) b. Out Range: 13% (24) Results
• Source: Pink Noise • Ground fact angle: 65o • Measured angles: (201) 1. Mean: 54.1o 2. STD: 3o 3. Error: 10o 4. Robustness: 100% (Target Range = ±5) 5. Range: 11o a. In Range: 98% (198) b. Out Range: 2% (3) Results
• Source: Pink Noise • Ground fact angle: 70o • Measured angles: (190) 1. Mean: 58.9o 2. STD: 3.25o 3. Error: 11o 4. Robustness: 100% (Target Range = ±5) 5. Range: 13o a. In Range: 13% (24) b. Out Range: 87% (166) Results
• Source: Pink Noise • Ground fact angle: 75o • Measured angles: (185) 1. Mean: 65.9o 2. STD: 3.5o 3. Error: 9o 4. Robustness: 100% (Target Range = ±5) 5. Range: 21o a. In Range: 5% (8) b. Out Range: 95% (177) Results
• Source: Pink Noise • Ground fact angle: 80o • Measured angles: (97) 1. Mean: 73.9o 2. STD: 0.38o 3. Error: 6o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2o a. In Range: 2.5% (2) b. Out Range: 97.5% (95) Results
• Source: Pink Noise • Ground fact angle: 85o • Measured angles: (195) 1. Mean: 71.4o 2. STD: 7.9o 3. Error: 8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 30o a. In Range: 15% (30) b. Out Range: 85% (165) Results
• Source: Pink Noise • Ground fact angle: 90o • Measured angles: (88) 1. Mean: 79.83o 2. STD: 3.19o 3. Error: 10o 4. Robustness: 100% (Target Range = ±5) 5. Range: 25o a. In Range: 13% (11) b. Out Range: 87% (77) Results
• Source: Pink Noise • Ground fact angle: -5o • Measured angles: (136) 1. Mean: -5.07o 2. STD: 0.11o 3. Error: 0.07o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.67o a. In Range: 100% (136) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -10o • Measured angles: (87) 1. Mean: -10.19o 2. STD: 0.13o 3. Error: 0.2o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.6o a. In Range: 100% (87) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -15o • Measured angles: (124) 1. Mean: -15.36o 2. STD: 0.16o 3. Error: 0.35o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.8o a. In Range: 100% (124) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -20o • Measured angles: (200) 1. Mean: -20.4o 2. STD: 0.2o 3. Error: 0.4o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.1o a. In Range: 100% (200) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -25o • Measured angles: (200) 1. Mean: -25.3o 2. STD: 0.23o 3. Error: 0.3o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.3o a. In Range: 100% (200) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -30o • Measured angles: (199) 1. Mean: -30.4o 2. STD: 0.29o 3. Error: 0.44o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.5o a. In Range: 100% (199) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -35o • Measured angles: (200) 1. Mean: -37.16o 2. STD: 0.33o 3. Error: 2.1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2.5o a. In Range: 100% (200) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -40o • Measured angles: (200) 1. Mean: -42.3o 2. STD: 0.39o 3. Error: 2.3o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2.4o a. In Range: 100% (200) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -45o • Measured angles: (200) 1. Mean: -47.74o 2. STD: 0.47o 3. Error: 2.73o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2.9o a. In Range: 100% (200) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -50o • Measured angles: (201) 1. Mean: -48.8o 2. STD: 0.6o 3. Error: 1.1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 3.8o a. In Range: 100% (201) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -55o • Measured angles: (199) 1. Mean: -57.5o 2. STD: 0.76o 3. Error: 2.1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 4.6o a. In Range: 100% (201) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -60o • Measured angles: (200) 1. Mean: -61.52o 2. STD: 0.96o 3. Error: 1.5o 4. Robustness: 100% (Target Range = ±5) 5. Range: 5.4o a. In Range: 100% (200) b. Out Range: 0% (0) Results
• Source: Pink Noise • Ground fact angle: -65o • Measured angles: (199) 1. Mean: -66.5o 2. STD: 1.3o 3. Error: 1.5o 4. Robustness: 100% (Target Range = ±5) 5. Range: 8.7o a. In Range: 87% (174) b. Out Range: 13% (25) Results
• Source: Pink Noise • Ground fact angle: -70o • Measured angles: (198) 1. Mean: -71o 2. STD: 2.6o 3. Error: 1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 18o a. In Range: 91% (182) b. Out Range: 9% (16) Results
• Source: Pink Noise • Ground fact angle: -75o • Measured angles: (159) 1. Mean: -72.1o 2. STD: 8.5o 3. Error: 2.8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 37o a. In Range: 84% (134) b. Out Range: 16% (25) Results
• Source: Pink Noise • Ground fact angle: -80o • Measured angles: (201) 1. Mean: -71o 2. STD: 11.2o 3. Error: 8.8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 43o a. In Range: 44% (90) b. Out Range: 56% (111) Results
• Source: Pink Noise • Ground fact angle: -85o • Measured angles: (201) 1. Mean: -63.7o 2. STD: 11.7o 3. Error: 21o 4. Robustness: 100% (Target Range = ±5) 5. Range: 44o a. In Range: 16% (32) b. Out Range: 84% (169) Results
• Source: Pink Noise • Ground fact angle: -90o • Measured angles: (200) 1. Mean: -60.8o 2. STD: 10o 3. Error: 29o 4. Robustness: 100% (Target Range = ±5) 5. Range: 48o a. In Range: 2% (3) b. Out Range: 98% (197) Results
ANALYSIS AND DISCUSSION • Source: Pink Noise • Ground fact angle: -90o to + 90o • Measured duration: 10 to 20 sec 1. Pink Source 2. Speech 3. Anechoic Speech Statistics • Mean • STD • Error w.r.t. Ground angles • Range (Target Range = ±5 @ Kinect) • Stability / Robustness • In Range: % • Out Range: %
ANALYSIS AND DISCUSSION Mean Angle STD Measurement Range Error STD
ANALYSIS AND DISCUSSION Mean Angle STD Measurement Range Error
ANALYSIS AND DISCUSSION In/Out Bound In/Out Bound

More Related Content

PDF
20211118 AI+ Remote Sensing
Jui-Hsin (Larry) Lai
 
PDF
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
Jui-Hsin (Larry) Lai
 
PDF
Object Tracking with Instance Matching and Online Learning
Jui-Hsin (Larry) Lai
 
PPTX
Sound Source Localization
Muhammad Imran
 
PDF
Sound Source Localization with microphone arrays
Ramin Anushiravani
 
PDF
Food Preservation for Farming Communities in Nepal: A Low Cost Engineering So...
Shristi Pradhan
 
PDF
FPGA Based Acoustic Source Localization Project
Shristi Pradhan
 
PDF
Immersive audio rendering for interactive complex virtual architectural envir...
Muhammad Imran
 
20211118 AI+ Remote Sensing
Jui-Hsin (Larry) Lai
 
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
Jui-Hsin (Larry) Lai
 
Object Tracking with Instance Matching and Online Learning
Jui-Hsin (Larry) Lai
 
Sound Source Localization
Muhammad Imran
 
Sound Source Localization with microphone arrays
Ramin Anushiravani
 
Food Preservation for Farming Communities in Nepal: A Low Cost Engineering So...
Shristi Pradhan
 
FPGA Based Acoustic Source Localization Project
Shristi Pradhan
 
Immersive audio rendering for interactive complex virtual architectural envir...
Muhammad Imran
 

Similar to Kinect Microphone Array case study (20)

PDF
2+3D Photography 2017 – INV 7 The 3D Image Capture Moonshot: Managing the Ene...
rijksmuseum
 
PDF
Coherent Adaptive Focusing Technology for the Inspection of Complex Geometry
Olympus IMS
 
PDF
NTEK Dodecahedral Sound Source
Kvikoo
 
PDF
NTEK Omni-directional Sound Source (5" 3D)
Kvikoo
 
PDF
NTEK Omni-directional Sound Source (4" HP)
Kvikoo
 
PDF
Advantages of the LA-950 Laser Diffraction PSA
HORIBA Particle
 
PDF
An Efficient Approach to Extract Singular Points for Fingerprint Recognition
Mesbah Uddin Khan
 
PPTX
Calibration of Phased array UT and TOFD (1).pptx
aravind449680
 
PDF
NTEK Omni-directional Sound Source (4" LT)
Kvikoo
 
PPTX
Entropy-based Histograms for Selectivity Estimation
University of Southern California
 
PPSX
Exploring Methods to Improve Edge Detection with Canny Algorithm
Prasad Thakur
 
PDF
Hvordan maler og dokumenterer du?
Plastindustrien
 
PDF
IRDC Theory and practice of building instruments for diffuse reflectance meas...
Bob Schumann
 
PDF
Selection of ear defenders
Mike Slater
 
PDF
20Dieterich-SRSSRTDosimetry.pdf
Nishant835443
 
PDF
NTEK Directional Loudspeaker
Kvikoo
 
PDF
19Mackie-IMRTDosimetry.pdf
Nishant835443
 
PDF
hh2014_presentation2noBackup
Jere Harrison
 
PDF
lesson 2 digital data acquisition and data processing
Mathew John
 
PPT
Using nasal curves matching for expression robust 3D nose recognition
Mehryar (Mike) E., Ph.D.
 
2+3D Photography 2017 – INV 7 The 3D Image Capture Moonshot: Managing the Ene...
rijksmuseum
 
Coherent Adaptive Focusing Technology for the Inspection of Complex Geometry
Olympus IMS
 
NTEK Dodecahedral Sound Source
Kvikoo
 
NTEK Omni-directional Sound Source (5" 3D)
Kvikoo
 
NTEK Omni-directional Sound Source (4" HP)
Kvikoo
 
Advantages of the LA-950 Laser Diffraction PSA
HORIBA Particle
 
An Efficient Approach to Extract Singular Points for Fingerprint Recognition
Mesbah Uddin Khan
 
Calibration of Phased array UT and TOFD (1).pptx
aravind449680
 
NTEK Omni-directional Sound Source (4" LT)
Kvikoo
 
Entropy-based Histograms for Selectivity Estimation
University of Southern California
 
Exploring Methods to Improve Edge Detection with Canny Algorithm
Prasad Thakur
 
Hvordan maler og dokumenterer du?
Plastindustrien
 
IRDC Theory and practice of building instruments for diffuse reflectance meas...
Bob Schumann
 
Selection of ear defenders
Mike Slater
 
20Dieterich-SRSSRTDosimetry.pdf
Nishant835443
 
NTEK Directional Loudspeaker
Kvikoo
 
19Mackie-IMRTDosimetry.pdf
Nishant835443
 
hh2014_presentation2noBackup
Jere Harrison
 
lesson 2 digital data acquisition and data processing
Mathew John
 
Using nasal curves matching for expression robust 3D nose recognition
Mehryar (Mike) E., Ph.D.
 
Ad

Recently uploaded (20)

PPTX
Safety Seminar civil to be ensured for safe working.
DILJEETKUMAR8
 
PPTX
Artificial Intelligence
dikshendrasarpate2
 
PDF
SMART SIGNAL TIMING FOR URBAN INTERSECTIONS USING REAL-TIME VEHICLE DETECTI...
Niraj Aarya
 
PDF
A SYSTEMATIC REVIEW OF APPLICATIONS IN FRAUD DETECTION
IJNSA Journal
 
PDF
Human-AI Collaboration: Balancing Agentic AI and Autonomy in Hybrid Systems
ijccsa
 
PPT
Occupational Health and Safety Management System
Akshay Kant Mishra
 
PDF
Visual Aids for Exploratory Data Analysis.pdf
Ashutosh Satapathy
 
PDF
COURSE DESCRIPTOR OF SURVEYING R24 SYLLABUS
MNANDITHACIVILSTAFF
 
PDF
Influence of Green Infrastructure on Residents’ Endorsement of the New Ecolog...
Journal of Contemporary Urban Affairs
 
PPTX
Current and future trends in Computer Vision.pptx
Subramanyam Natarajan
 
PPTX
Graph Data Structures with Types, Traversals, Connectivity, and Real-Life App...
usham61
 
PPTX
Sorting and Hashing in Data Structures with Algorithms, Techniques, Implement...
usham61
 
PDF
BIO-INSPIRED ARCHITECTURE FOR PARSIMONIOUS CONVERSATIONAL INTELLIGENCE : THE ...
ijaia
 
PPT
INTRODUCTION -Data Warehousing and Mining-M.Tech- VTU.ppt
Subramanyam Natarajan
 
PDF
null (2) bgfbg bfgb bfgb fbfg bfbgf b.pdf
Ramez Hosny
 
PDF
Level 2 – IBM Data and AI Fundamentals (1)_v1.1.PDF
KSRIHARISASANKA
 
PDF
Soil Improvement Techniques Note - Rabbi
rahmatideal000
 
PPTX
CURRICULAM DESIGN engineering FOR CSE 2025.pptx
Amuda3
 
PDF
Categorization of Factors Affecting Classification Algorithms Selection
IJDKP
 
PDF
UNIT no 1 INTRODUCTION TO DBMS NOTES.pdf
pawarbhaktiit
 
Safety Seminar civil to be ensured for safe working.
DILJEETKUMAR8
 
Artificial Intelligence
dikshendrasarpate2
 
SMART SIGNAL TIMING FOR URBAN INTERSECTIONS USING REAL-TIME VEHICLE DETECTI...
Niraj Aarya
 
A SYSTEMATIC REVIEW OF APPLICATIONS IN FRAUD DETECTION
IJNSA Journal
 
Human-AI Collaboration: Balancing Agentic AI and Autonomy in Hybrid Systems
ijccsa
 
Occupational Health and Safety Management System
Akshay Kant Mishra
 
Visual Aids for Exploratory Data Analysis.pdf
Ashutosh Satapathy
 
COURSE DESCRIPTOR OF SURVEYING R24 SYLLABUS
MNANDITHACIVILSTAFF
 
Influence of Green Infrastructure on Residents’ Endorsement of the New Ecolog...
Journal of Contemporary Urban Affairs
 
Current and future trends in Computer Vision.pptx
Subramanyam Natarajan
 
Graph Data Structures with Types, Traversals, Connectivity, and Real-Life App...
usham61
 
Sorting and Hashing in Data Structures with Algorithms, Techniques, Implement...
usham61
 
BIO-INSPIRED ARCHITECTURE FOR PARSIMONIOUS CONVERSATIONAL INTELLIGENCE : THE ...
ijaia
 
INTRODUCTION -Data Warehousing and Mining-M.Tech- VTU.ppt
Subramanyam Natarajan
 
null (2) bgfbg bfgb bfgb fbfg bfbgf b.pdf
Ramez Hosny
 
Level 2 – IBM Data and AI Fundamentals (1)_v1.1.PDF
KSRIHARISASANKA
 
Soil Improvement Techniques Note - Rabbi
rahmatideal000
 
CURRICULAM DESIGN engineering FOR CSE 2025.pptx
Amuda3
 
Categorization of Factors Affecting Classification Algorithms Selection
IJDKP
 
UNIT no 1 INTRODUCTION TO DBMS NOTES.pdf
pawarbhaktiit
 
Ad

Kinect Microphone Array case study

  • 1. PROCEDURE FOR LOCALIZATION AND ARRAYS DESIGNING KINECT MICROPHONE ARRAY CASE STUDY FOR LOCALIZATION PERFORMANCE 2016-10-08 MUHAMMAD IMRAN
  • 2. CONTENTS • Measurement Procedures (Data Acquisition) • Array Performance Analysis • Localization Performance • Kinect Case study for Localization Performance
  • 4. MEASUREMENT PROCEDURES (DATA ACQUISITION) 1. Measurement Setup a. Data Acquisition HWD b. Source Selection • Wideband Source • Narrowband Source (Single Frequency) • Speech Source c. Source Directivity • Narrow (±20) • Speech Source d. Geometry Plan • Placement of source and receiver • Alignment of Array
  • 5. DATA ACQUISITION HARDWARE a. Genelec 8030 • 20 seconds • Sampling = 44.1 KHz (Prefer) • 860” Frames (if Frame size is 1024) • 860” DOAs for One position b. Kinect Array • 04-CH: Linear Array • Non-Uniform • Windows Direct Sound • Interface • Source: Professor Edgar Choueiri, 3D3A Lab, Princeton University • For Details: Microsoft Corporation
  • 6. PLACEMENT AND ALIGNMENT • Making Grid (Physical) • Thread Network • Protractor • Separation between elements • Error within 1 mm • Axial Placement • Alignment Errors • Within the limits of (Hz, Vt) • -0.5o to +0.5o • Environmental Factors • Temperature (STP) • Humidity (STP) • Back Ground Noise Level <= 20dB (±5dB)1
  • 7. PROCEDURE FOR DATA ACQUISITION • Kinect alignment to Loudspeaker • With Magnetic Compass • Exciting Sound Sources: 1. Pink Noise (Wideband Source) • Duration = 10 sec • Measurement Angles = -90 to + 90 (Step: 05o) • For Linear Array 2. Sine 500, 1000, 1500 Hz • Duration = 10 sec • Radom angles (05-10 angles) 3. Speech (Anechoic) • Duration = 10 sec • Measurement Angles = -90 to + 90 (10o)
  • 8. ARRAY DIRECTIVITY • Figure 1: (Ref: 02): Microphone array directivity pattern in three dimensions • What to measure (Ref: 01) (Formulae and Matlab Example Code) 1. Microphone Sensitivity 2. Microphone Noise and Output SNR 3. Array Directivity Pattern 4. Array Frequency Response 5. Array Directivity Index Figure 1 Reference: 1. Chapter: 03: Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009 2. MIT Media Lab: Kinect Array Directivity
  • 9. ARRAY DIRECTIVITY • Figure 2: A directivity pattern of Kinect (Azimuth @ 0o Elev.) Figure 2a Figure 2b Reference: 1. Chapter: 03, Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009
  • 10. DIRECTIVITY INDEX COMPARISON • Figure 3: A directivity Index of different number of microphone • More Microphones more directivity • More computational cost • Less Microphones Less directivity • Less computational cost Directivity Index 0 2 4 6 8 10 12 Uni-directional Tw o element Four element Microphone/Mic Array DI,dB Figure 3 Reference: 1. Chapter: 05, Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009
  • 12. PERFORMANCE ANALYSIS PROCEDURES (DOA) 1. Flow of computation a. Data segmentation b. Frame by Frame Computation a. Localization Algo (GCC-PHAT) b. Individual Pair Computing c. Combining Pairs 2. Verification Parameters a. Localization Accuracy b. Localization Resolution c. Localization Stability (Standard deviation) (The standard deviation tells the dispersion of data) d. Robustness (Root Mean Squared Error: RMS) 3. Plotting and Describing Results a. Histograms, Polar Plots
  • 13. FLOW OF COMPUTATION Flow of computation a. Data segmentation b. Frame by Frame Computation a. GCC-PHAT b. Individual Pair Computing c. Combining Pairs c. Recorded signal Length a. 10 sec d. Sampling a. 44100 Samples/Sec e. Total Frames a. 430 (@ 1024) f. Total DOA a. 430 g. Recommended minimum = 200 CH1 CH2 CH3 CH4 Signal Length (10 sec)
  • 14. VERIFICATION PARAMETERS 1. Plotting and Describing Results a. Histograms, Polar Plots 2. Verification Parameters 1. Localization Resolution 2. *Localization Accuracy (Error) 3. *Localization Stability (Standard deviation) 4. **Confidence Level Reference: *Chapter: 06, Tashev, Ivan Jelev. Sound capture and processing: practical approaches. John Wiley & Sons, 2009 **https://en.wikipedia.org/wiki/Confidence_interval
  • 16. MEASUREMENTS • Source: Pink Noise • Ground fact angle: -90o to + 90o • Measured duration: 10 to 20 sec 1. Pink Source 2. Speech 3. Anechoic Speech Statistics • Mean • STD • Error w.r.t. Ground angles • Range (Target Range = ±5 @ Kinect) • Stability / Robustness • In Range: % • Out Range: %
  • 17. • Source: Pink Noise • Ground fact angle: 0o • Measured angles: (533) 1. Mean: -0.12o 2. STD: 0.11o 3. Error: 0.13o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.6o a. In Range: 100% (533) b. Out Range: 0% (0) Results
  • 18. • Source: Pink Noise • Ground fact angle: 5o • Measured angles: (533) 1. Mean: 4.97o 2. STD: 0.18o 3. Error: 0.028o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1o a. In Range: 100% (533) b. Out Range: 0% (0) Results
  • 19. • Source: Pink Noise • Ground fact angle: 10o • Measured angles: (543) 1. Mean: 9.87o 2. STD: 0.134o 3. Error: 0.13o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.6o a. In Range: 100% (543) b. Out Range: 0% (0) Results
  • 20. • Source: Pink Noise • Ground fact angle: 15o • Measured angles: (133) 1. Mean: 15o 2. STD: 0.25o 3. Error: 0.004o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.3o a. In Range: 100% (133) b. Out Range: 0% (0) Results
  • 21. • Source: Pink Noise • Ground fact angle: 20o • Measured angles: (533) 1. Mean: 20.16o 2. STD: 0.19o 3. Error: 0.15o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.9o a. In Range: 100% (533) b. Out Range: 0% (0) Results
  • 22. • Source: Pink Noise • Ground fact angle: 25o • Measured angles: (541) 1. Mean: 25o 2. STD: 0.25o 3. Error: 0.001o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.4o a. In Range: 100% (541) b. Out Range: 0% (0) Results
  • 23. • Source: Pink Noise • Ground fact angle: 30o • Measured angles: (534) 1. Mean: 29.7o 2. STD: 0.17o 3. Error: 0.29o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.8o a. In Range: 100% (534) b. Out Range: 0% (0) Results
  • 24. • Source: Pink Noise • Ground fact angle: 35o • Measured angles: (519) 1. Mean: 34.05o 2. STD: 0.17o 3. Error: 0.95o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1o a. In Range: 100% (526) b. Out Range: 0% (0) Results
  • 25. • Source: Pink Noise • Ground fact angle: 40o • Measured angles: (526) 1. Mean: 39.08o 2. STD: 0.19o 3. Error: 0.91o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1o a. In Range: 100% (526) b. Out Range: 0% (0) Results
  • 26. • Source: Pink Noise • Ground fact angle: 45o • Measured angles: (190) 1. Mean: 43.9o 2. STD: 0.37o 3. Error: 2.9 4. Robustness: 100% (Target Range = ±5) 5. Range: 3o a. In Range: 100% (190) b. Out Range: 0% (0) Results
  • 27. • Source: Pink Noise • Ground fact angle: 50o • Measured angles: (215) 1. Mean: 48.02o 2. STD: 0.5o 3. Error: 2o 4. Robustness: 100% (Target Range = ±5) 5. Range: 3o a. In Range: 100% (215) b. Out Range: 0% (0) Results
  • 28. • Source: Pink Noise • Ground fact angle: 55o • Measured angles: (195) 1. Mean: 54.1o 2. STD: 1.5o 3. Error: 0.8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 7o a. In Range: 94% (204) b. Out Range: 6% (11) Results
  • 29. • Source: Pink Noise • Ground fact angle: 60o • Measured angles: (200) 1. Mean: 58.2o 2. STD: 2.34o 3. Error: 1.7o 4. Robustness: 100% (Target Range = ±5) 5. Range: 8.9o a. In Range: 87% (176) b. Out Range: 13% (24) Results
  • 30. • Source: Pink Noise • Ground fact angle: 65o • Measured angles: (201) 1. Mean: 54.1o 2. STD: 3o 3. Error: 10o 4. Robustness: 100% (Target Range = ±5) 5. Range: 11o a. In Range: 98% (198) b. Out Range: 2% (3) Results
  • 31. • Source: Pink Noise • Ground fact angle: 70o • Measured angles: (190) 1. Mean: 58.9o 2. STD: 3.25o 3. Error: 11o 4. Robustness: 100% (Target Range = ±5) 5. Range: 13o a. In Range: 13% (24) b. Out Range: 87% (166) Results
  • 32. • Source: Pink Noise • Ground fact angle: 75o • Measured angles: (185) 1. Mean: 65.9o 2. STD: 3.5o 3. Error: 9o 4. Robustness: 100% (Target Range = ±5) 5. Range: 21o a. In Range: 5% (8) b. Out Range: 95% (177) Results
  • 33. • Source: Pink Noise • Ground fact angle: 80o • Measured angles: (97) 1. Mean: 73.9o 2. STD: 0.38o 3. Error: 6o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2o a. In Range: 2.5% (2) b. Out Range: 97.5% (95) Results
  • 34. • Source: Pink Noise • Ground fact angle: 85o • Measured angles: (195) 1. Mean: 71.4o 2. STD: 7.9o 3. Error: 8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 30o a. In Range: 15% (30) b. Out Range: 85% (165) Results
  • 35. • Source: Pink Noise • Ground fact angle: 90o • Measured angles: (88) 1. Mean: 79.83o 2. STD: 3.19o 3. Error: 10o 4. Robustness: 100% (Target Range = ±5) 5. Range: 25o a. In Range: 13% (11) b. Out Range: 87% (77) Results
  • 36. • Source: Pink Noise • Ground fact angle: -5o • Measured angles: (136) 1. Mean: -5.07o 2. STD: 0.11o 3. Error: 0.07o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.67o a. In Range: 100% (136) b. Out Range: 0% (0) Results
  • 37. • Source: Pink Noise • Ground fact angle: -10o • Measured angles: (87) 1. Mean: -10.19o 2. STD: 0.13o 3. Error: 0.2o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.6o a. In Range: 100% (87) b. Out Range: 0% (0) Results
  • 38. • Source: Pink Noise • Ground fact angle: -15o • Measured angles: (124) 1. Mean: -15.36o 2. STD: 0.16o 3. Error: 0.35o 4. Robustness: 100% (Target Range = ±5) 5. Range: 0.8o a. In Range: 100% (124) b. Out Range: 0% (0) Results
  • 39. • Source: Pink Noise • Ground fact angle: -20o • Measured angles: (200) 1. Mean: -20.4o 2. STD: 0.2o 3. Error: 0.4o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.1o a. In Range: 100% (200) b. Out Range: 0% (0) Results
  • 40. • Source: Pink Noise • Ground fact angle: -25o • Measured angles: (200) 1. Mean: -25.3o 2. STD: 0.23o 3. Error: 0.3o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.3o a. In Range: 100% (200) b. Out Range: 0% (0) Results
  • 41. • Source: Pink Noise • Ground fact angle: -30o • Measured angles: (199) 1. Mean: -30.4o 2. STD: 0.29o 3. Error: 0.44o 4. Robustness: 100% (Target Range = ±5) 5. Range: 1.5o a. In Range: 100% (199) b. Out Range: 0% (0) Results
  • 42. • Source: Pink Noise • Ground fact angle: -35o • Measured angles: (200) 1. Mean: -37.16o 2. STD: 0.33o 3. Error: 2.1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2.5o a. In Range: 100% (200) b. Out Range: 0% (0) Results
  • 43. • Source: Pink Noise • Ground fact angle: -40o • Measured angles: (200) 1. Mean: -42.3o 2. STD: 0.39o 3. Error: 2.3o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2.4o a. In Range: 100% (200) b. Out Range: 0% (0) Results
  • 44. • Source: Pink Noise • Ground fact angle: -45o • Measured angles: (200) 1. Mean: -47.74o 2. STD: 0.47o 3. Error: 2.73o 4. Robustness: 100% (Target Range = ±5) 5. Range: 2.9o a. In Range: 100% (200) b. Out Range: 0% (0) Results
  • 45. • Source: Pink Noise • Ground fact angle: -50o • Measured angles: (201) 1. Mean: -48.8o 2. STD: 0.6o 3. Error: 1.1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 3.8o a. In Range: 100% (201) b. Out Range: 0% (0) Results
  • 46. • Source: Pink Noise • Ground fact angle: -55o • Measured angles: (199) 1. Mean: -57.5o 2. STD: 0.76o 3. Error: 2.1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 4.6o a. In Range: 100% (201) b. Out Range: 0% (0) Results
  • 47. • Source: Pink Noise • Ground fact angle: -60o • Measured angles: (200) 1. Mean: -61.52o 2. STD: 0.96o 3. Error: 1.5o 4. Robustness: 100% (Target Range = ±5) 5. Range: 5.4o a. In Range: 100% (200) b. Out Range: 0% (0) Results
  • 48. • Source: Pink Noise • Ground fact angle: -65o • Measured angles: (199) 1. Mean: -66.5o 2. STD: 1.3o 3. Error: 1.5o 4. Robustness: 100% (Target Range = ±5) 5. Range: 8.7o a. In Range: 87% (174) b. Out Range: 13% (25) Results
  • 49. • Source: Pink Noise • Ground fact angle: -70o • Measured angles: (198) 1. Mean: -71o 2. STD: 2.6o 3. Error: 1o 4. Robustness: 100% (Target Range = ±5) 5. Range: 18o a. In Range: 91% (182) b. Out Range: 9% (16) Results
  • 50. • Source: Pink Noise • Ground fact angle: -75o • Measured angles: (159) 1. Mean: -72.1o 2. STD: 8.5o 3. Error: 2.8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 37o a. In Range: 84% (134) b. Out Range: 16% (25) Results
  • 51. • Source: Pink Noise • Ground fact angle: -80o • Measured angles: (201) 1. Mean: -71o 2. STD: 11.2o 3. Error: 8.8o 4. Robustness: 100% (Target Range = ±5) 5. Range: 43o a. In Range: 44% (90) b. Out Range: 56% (111) Results
  • 52. • Source: Pink Noise • Ground fact angle: -85o • Measured angles: (201) 1. Mean: -63.7o 2. STD: 11.7o 3. Error: 21o 4. Robustness: 100% (Target Range = ±5) 5. Range: 44o a. In Range: 16% (32) b. Out Range: 84% (169) Results
  • 53. • Source: Pink Noise • Ground fact angle: -90o • Measured angles: (200) 1. Mean: -60.8o 2. STD: 10o 3. Error: 29o 4. Robustness: 100% (Target Range = ±5) 5. Range: 48o a. In Range: 2% (3) b. Out Range: 98% (197) Results
  • 54. ANALYSIS AND DISCUSSION • Source: Pink Noise • Ground fact angle: -90o to + 90o • Measured duration: 10 to 20 sec 1. Pink Source 2. Speech 3. Anechoic Speech Statistics • Mean • STD • Error w.r.t. Ground angles • Range (Target Range = ±5 @ Kinect) • Stability / Robustness • In Range: % • Out Range: %
  • 55. ANALYSIS AND DISCUSSION Mean Angle STD Measurement Range Error STD
  • 56. ANALYSIS AND DISCUSSION Mean Angle STD Measurement Range Error
  • 57. ANALYSIS AND DISCUSSION In/Out Bound In/Out Bound