Improved 3D Sound Deliveredto Headphones Using Wavelets

By

Ozlem KALINLI

EE-SystemsUniversity of Southern California

December 4, 2003

Outline:

Introduction

Work

Results

Conclusion

Immersive Audio Environments Transport listener into the same sonic environment as the event

o Multiple, spatially-distributed

sound sourceso Head and source motiono Room Acoustics

Virtually listening environmentso Synthetic acoustic images (headphones or loudspeakers)o Simulated directional sound informationo Simulated room acoustics

Introduction

Immersive Reproduction of 3D Sound Scheme

Head Related Transfer Function (HRTF)

Head Related Transfer Function (HRTF)

Special transformation of a source from a point in free space to the listener’s eardrums.

HRTF measurements are computed using a dummy head (KEMAR)

Used for sound localizationSound Transmission from Source to Listener

Introduction

Sound Localization

Localization of sound, cues:o Interaural time difference

(ITD) , dominant below 1.5 kHz

o Interaural intensity difference (IID), dominant above 3 kHz

Reasons:o Path length differenceo Head Shadowingo Reflection of Head

Introduction

Main Work Goal of Work: To obtain a better sound diffusion

from the mono-sound recorded at an anechoic chamber

System Toolso Use HRTF to localize

sound, 30o azimuth, and 0o elevationo Use wavelet filter banks with

time delay at the lowest frequency (below 1.5 kHz) to get the sound diffusion (adding reverberant sound)

Work

Overall System

Work

• Fs= 44.1 kHz, 16 bit

• 5 Stages of dyadic tree to get the signal below 1.5 kHz

• Daubechies wavelets, with filter tap 16

• Delay time 7.25 ms

Simulation Results 4 different types of audio signals are tested Piano, guitar, classical music, pop song

Time Domain Waveforms for Piano Sound (Left Channel)

(a) HRTF Sound (b) Delayed Sound with Wavelet (c) Final Sound

Results

Results for Piano Sound Subjective Listening Tests

Relation Between Time Delays and Correlation Coefficient

0

1

2

3

4

5

6

people

More Diffusive

Listening Tests

HRTF Sound

Delayed Sound

Final Sound

0

1

2

3

4

people

Sounds Better

Listening Tests

HRTF Sound

Delayed Sound

Final Sound

Time Delay [ms]

Correlation Coefficient

Delayed Sound Final Sound

7.25 -0.3994 0.3577

14.5 -0.3235 0.3002

17.4 -0.3566 0.3377

Results

Other Work Done

Sound localized at 110o of azimuth with 0o elevation is also tested, since surround sound is desired at the + 110o and - 110o

o Listening test results similar to the 30o of azimuth

o Relation Between Time Delays and Correlation Coefficient

Time Delay [ms]

Correlation Coefficient

Delayed Sound Final Sound

7.25 -0.2905 -0.0803

14.5 -0.2024 -0.1194

17.4 -0.2894 -0.1254

Results

Results for Piano Sound

Original sound, Mono HRTF-30

o Test signal (no delay)o Delayed Sound (7.25 ms)o Final Sound

HRTF-110o Delayed Sound (7.25 ms)o Final Sound

7.25 ms 14.5 ms 17.4 ms

7.25 ms 14.5 ms 17.4 ms

Results

Conclusion Introducing delay in the frequency band

below 1.5 kHz produces reverberant sound

The final sound is better than HRTF sound in sense of the sound diffusion.

Depending on the audio characteristic, the optimum delay time to obtain de-correlated sound (small correlation coefficient) may vary.

When the delay is very high, it simulates big halls.

Conclusion

References “Improved 3D Sound Using Wavelets”, U. P. Chong, H.

Kim, K. N. Kim, IEEE Information Systems and Technologies, 2001.

“HRTF Measurements of a KEMAR Dummy-Head Microphone”, MIT Media Lab Perceptual Computing- Technical Report #280.

“HRTF Measurements of a KEMAR Dummy-Head Microphone”, http://sound.media.mit.edu/KEMAR.html

“Virtually Auditory Space Generation and Applications”, Simon Carlie, Chapman and Hall, 1996.