1. Microphone Array Project ECE5525 – Speech Processing Robert Villmow 12/11/03

2. Microphone Array Purpose Improves sound reception when interfering sources are located far enough apart spatial separation Capable of determinig the direction a sound is coming from

3. Background Data sampled at each microphone is different due to different angles of arrival Delay and Sum Algorithm Assume a sound is arriving at a specific angle then calculate when that sound will be sampled by the other microphones Sound arriving from desired angle will be correlated between microphones Sound arriving from other angles will be uncorrelated Summation of the data will attenuate uncorrelated signals

4. 15 Element Microphone Array High Middle Low Low Array 4x spacing as high array Middle Array 2x spacing as high array High Array minimum spacing between microphones

5. Microphone Sampling Source at 10° Source at 50° Source at 90° Source at 0°

6. Sampling Summary Sample rate affects performance sampled data differs between microphones algorithm depends on data being correlated between microphonesS Sound sources at 0° are not affected by sample rate sampled by all microphones at the same time

7. High Array Low Frequency Response Similar to a single microphone High Frequency Response Narrow beam width Frequency Responses at 4400 Hz

8. Low Array Low Frequency Response Good Directional response High Frequency Response Too many Grating lobes Frequency Responses at 800 Hz

9. Block Diagram BP Filter High Freq BP Filter Middle Freq BPFilter Low Freq 2.5khz – 4.5khz 1.5khz – 2.5khz 300hz – 1.5khz    

10. Array Test Configuration Speaker Radio 45°

11. CMU Microphone Array Data 15 Channels 16 kHz, 16-bit linear Sampling Files used an101-mtms-arr3A.adc 3 cm spacing in a noisy computer lab Speaker is 1 meter from array an101-mtms-arr4A.adc 4 cm spacing in a noisy computer lab Speaker is 1 meter from array an101-mtms-arrCR1A.adc 4 cm spacing with radio at 45° Speaker at a distance of 1 meters

12. Arr3A Results Difference – Single Channel vs. Summed signal Mean Squared Error – Single Channel vs. Summed signal High Middle Low

13. Arr3A – Mean Squared Error ArrayMaxMin High9.8070 x 10 6 3.3647 x 10 7 Middle1.0517 x 10 7 3.4093 x 10 7 Low7.7878 x 10 6 3.1970 x 10 7

14. Arr4A Results Mean Squared Error – Single Channel vs. Summed signal Difference – Single Channel vs. Summed signal High Middle Low

15. Arr4A – Mean Squared Error ArrayMaxMin High 8.1192 x 10 6 6.7789 x 10 7 Middle 8.3030 x 10 6 6.8715 x 10 7 Low 6.3974 x 10 6 6.7689 x 10 7

16. ArrCR1A Results (Interpolated) Difference – Single Channel vs. Summed signal Mean Squared Error – Single Channel vs. Summed signal

17. ArrCR1A – Mean Squared Error ArrayMaxMin High 5.0761 x 10 6 1.2176 x 10 7 Middle 5.2716 x 10 6 1.3283 x 10 7 Low 4.3089 x 10 6 1.3238 x 10 7

18. Sound Results Sound Files...

19. Summary Microphone array increases SNR Array Response is best at angles between ±30° Oversampling smoothes array response increased computations decreases performance lost due to sampling