Cancelling the Noise of Wall Street
How Noise Cancellation Works in a Trading Floor Environment
The harsh environment of trading floors, including the floor of the New York Stock Exchange, requires voice recognition solutions beyond that which the traditional recognizer can provide.
Research and development into new ways to make voice recognition work for Wall Street meant a new approach. The approach has been embodied in the Ume Voice, Inc. StreetLab. This article will discuss the initiatives which are being conducted in StreetLab, as well as describe the following present accomplishments and near term goals. The focus will be on the very recent development of the Noise Control Aperture in the StreetEars application.
Noise Control Aperture
The Noise Control Aperture (NCA) has been designed to enhance the performance of pressure differential microphones used to cancel or reject background noise. When both the microphone and the aperture are used together they form an electro-acoustic noise rejection system exceeding the performance of any other technology existing today.
Many microphone element designs employ front and rear sound ports which allow sound to enter both and impinge upon the diaphragm simultaneously in opposite directions, resulting in little or no signal being generated by the mic. This technique is applied in a wide variety of cardioid mics as well as telephone handset transmitters and headsets. Some employ acoustic tuning to the rear port to make it more frequency responsive.
Because of construction restraints inherent in microphone design, one port of the mic is always more sensitive. This results from the need to provide a supporting structure for the diaphragm and the resulting impedance that structure presents to sound entering the microphone element. In FIG 1 . the parts of an electret mic element are shown. The active electret material is located between the rear port and the diaphragm. This means sound entering from the rear port impinges on the electret before arriving at the diaphragm. Extra holes are punched in the back to help overcome the reduced sensitivity. In common practice, the more sensitive port is faced forward to capture the desired sound while the less sensitive port is utilized for capturing and nullifying the undesired background noises.
If the front and back sensitivities of the element were equal, then theoretically 100% noise rejection would be possible whenever noise of equal pressure is subjected to both entrances to the microphone. In practice however, only 10-20 dB noise reduction is possible using the currently available mic elements and this is only for frequencies below about 3kHz.
To overcome these limitations we have constructed the UmeVoice NCA, a flow control device. The NCA controls the flow of acoustic waves in such a way that the noise pressure of the rear of the diaphragm matches that on the front of the diaphragm. This phenomenon is illustrated in FIG 2a and b .
FIG 2a is a sketch without the flow control device. Here the noise pressure at the rear of the diaphragm is less than that at the front, as previously noted. This is accomplished using a pair of curved reflectors, a sound concentrator slot and a sound wall or bridge which separated two acoustic zones. FIG 2b illustrates the use of a flow control device.
Now the incident noise impinges upon the rear diaphragm from many more directions. Because the walls of the passive flow control device are curved, they present many angles of incidence for which some angles of reflection provide the correct path directing the noise to the slot and the rear of the mic element. Because the wall surfaces are large compared to the slot, a net gain in sound pressure is achieved within the slot and at the rear port of the mic element. This effect overcomes the higher impedance of the rear port, bringing the diaphragm pressure equilibrium nearer to null and increases the overall noise reduction of the system.
Microphones are tested for noise rejection by comparing each response to that of a Peavey ERO 10 reference microphone which has no noise rejection characteristics but exhibits well defined flat response from 20Hz to 20kHz. The reference mic and the test mic are placed in very close proximity to each other equidistant from the noise source FIG 3 .
A near field source of noise is provided by an acoustic dummy of human dimensions with a JBL Control Micro loudspeaker mounted inside the head.
Using a Hewlett Packard 3566 two channel dynamic spectrum analyzer for source noise and measurement, a white noise signal of 300mV is amplified (McGowan 354SL) and connected to the dummy loudspeaker and adjusted to 80 dB sound pressure at the mics.
The mics are routed to the analyzer through a Makie 1202 mixer with the reference mic to channel one and the test mic to channel two. With the analyzer in frequency response mode, the two signals are compared by the HP3566, which can automatically divide their power outputs.
After plotting the near field response, the amplifier is switched to the far field loudspeaker and without moving the mics, the sound pressure is again adjusted to 80dB at the mics. This requires turning up the amplifier volume because of the added distance between the loudspeaker and the mics. The far field response is then plotted to measure how much less responsive the mic is to distant sounds. The difference between the near field and the far field response is a measure of the microphones noise rejection.
In figure 4 , the upper trace is the near field response of the Knowles headset. Notice that it follows approximately the -10dB line. This means it has a fairly flat response but 10dB less gain than the reference mic. The lower trace represents the noise rejection of the mic, which varies between about 10 and 20 dB up to about 3.5 kHz.
From FIG 4 and FIG 5 it is clear that the noise rejection of the NCA is dramatically greater, ranging between 10 and 40 dB all the way up to 6.45kHz, as shown by the lower trace. These indicators show the NCA can provide a means of greatly enhancing the performance of existing microphone technology.
With the help of sophisticated noise cancellation, speech recognition can be effectively used even in the most frenzied environments, with the most demanding of customers, as its success on Wall Street has demonstrated.
Paul Cavise, Joe Tate and Narayanan Raju work for UmeVoice. For more information, contact Paul Cavise at: Ume Voice, Inc., (201) 938-0011, firstname.lastname@example.org., www.umevoice.com .