Performance
Our Performance Notes contain data and a number of recordings comparing the Acoustic Magic Voice Tracker™ to 1st generation array microphones in various, real life situations. You can download these recordings and hear the improved performance for yourself.
Automatic Speech Recognition Tests
Automatic speech recognition accuracy tests are important not only because they indicate how usable a microphone would be for dictation applications, but also because they provide a quantitative measure of microphone performance.
However, the error rate is dependent not only on the quality of the microphone, but also on the nature, location and intensity of the noise, the size and reverberant quality of the room, the clarity of the talker, and the microphone's capability to adjust to movement by the talker during dictation. Therefore, an error rate percentage should be analyzed carefully.
Below are several comparison plots for the Acoustic Magic Voice Tracker™ and two first generation array microphones. Each figure plots the accuracy scores (in %) vs. increasingly difficult conditions.
The first plot shows the more rapid increase in errors in 1st generation arrays as the distance from the array is increased. The second plot shows the more rapid increase in errors in 1st generation arrays as the talker moves off axis. The third plot shows the more rapid increase in errors in 1st generation arrays as the babble noise level is increased.
In all of these tests, the talker level was set at 85 dB, measured 5 inches from the talker. The noise was recorded office babble noise, and was placed 90 degrees off the talker axis, 6 feet from the array. The noise level was measured by averaging its power at the microphones over a period of 30 seconds. The room was a normal room in a home, 12 feet by 15 feet, with the array located at one end of the room, along the long wall. All scores are given as percentages of standard (a close talking microphone). For each data point in the plots, 2 trials of 1600 words each were performed and averaged. The ASR recognition engine was Dragon Naturally Speaking Preferred, Version 4.0. Note that for the second plot (ASR score vs. angle), the array has only 2 data points. This is due to the fact that for these angles, the array's signal-to-noise ratio was too low to produce reliable ASR scores.
Recordings
Recordings give an idea of the improved intelligibility (in the presence of noise) of the Acoustic Magic array microphone compared to a single unidirectional microphone and 1st generation array microphones.
In automatic speech recognition, this improved intelligibility leads to better recognition. In teleconferencing, it leads to better understanding at the listener end.
Acoustic Magic's array processing algorithms create better sound fidelity (less distortion and artifact) than 1st generation array microphones.
Acoustic Magic's spatial filtering and noise suppression algorithms do a more effective job of filtering out interference from other talkers or fan noise.
The first three recording sets show our microphone's ability to "hear" the talker in spite of interference from other talkers and fan noise.
- Recording set A: 8 feet, 20 degrees off axis, 65 dB fan noise
-
The speech volume is adjusted for peak level of 85 dB, measured 5" from the tripod mounted speaker. The background fan noise is played through a speaker located 90 degrees off axis, 6 feet from the array, at a level of 65 dB, measured at the array. The room was a normal room in a home, 12 feet by 15 feet, with the array located at one end of the room, along the long wall.
- Recording set B: 8 feet, 20 degrees off axis, 70 dB fan noise
-
This recording is the same setup as in A, but with more fan noise.
- Recording set C: Competing talkers
-
This recording substitutes a competing talker for the fan noise. The main talker is adjusted for a peak level of 85 dB, measured 5" from the tripod mounted speaker. The competing talker has a peak level of 65 dB measured at the array.
The next 2 recordings compare the Acoustic Magic array with two 1st generation arrays to illustrate our improved signal fidelity. In these recordings, babble noise was used rather than fan noise. The babble noise had a 30-second average level of 60 dB and was located 90 degrees off axis. This is the same environment in which the ASR tests were performed.
- Recording set D: 4 feet, 30 degrees off axis, 60 dB babble noise
- Recording set E: 8 feet, 20 degrees off axis, 60 dB babble noise
Video Conferencing
The Voice Tracker's™ noise cancellation technology, combined with it's long range and 180 degree scanning capability make it ideal for teleconferencing applications. The Voice Tracker™ can be placed on the monitor and still cover an entire conference room. The following video illustrates the Voice Tracker's™ ability to follow the talker around the room (or switch to other talkers). The talker starts 10 feet away from the microphone, then moves to the left, and then 25+ feet away. The room is highly reverberant, which makes the microphone's task more difficult. The recording was made using a Tandberg conferencing system.
Download Video Conferencing Demo (4.6 MB)
Next Generation Technology
Purchase the Voice Tracker Array Microphone in Our Online Store: Visit Our Online Store
