A new direction for IBM speech recognition: bionic bat ears can be accurately "listen" with sonar

The bat uses biological sonar to navigate the jungle in the night. Their ultrasonic pulses can position the sound more accurately than the artificial sonar device. To replicate and harness this capability, IBM Academy Award winner Rolf Müller designed a man-made bat ear in collaboration with his team at Virginia Tech.

Rolf Müller's research caught the attention of IBM. The neurological computing team of IBM expert Han Jinping, and IBM Watson voice expert Cui Xiaodong and his colleagues, saw Professor Müller's artificial "dynamic peripheral" (the bat's rotatable outer ear makes their biological sonar more accurate). ) the potential and hope to enhance the ability of human speech understanding. They included Müller's Ph.D. student Anupam Gupta in the team and together they explored the application of artificial bat bionic ears in speech processing.

They found that these bionic ears are not only very effective sonar devices, but also for speech recognition.

Artificial ear that mimics the bat

The research team copied a dynamic receiving system based on the ability of the bat to change the shape of the ear. It improves the accuracy of the Automatic Speech Recognition System (ASR) and more accurately locates the talker. Han Jinping will present this finding in their paper "Receiving dynamics inspired by the bud-headed bat to incorporate dynamic features into speech signals" and at the 172nd session of the American Acoustics Association this week.

These dynamic systems have the potential to evolve into voice-receiving devices that allow users to "listen like a bat." This will improve existing hearing aids and directional microphones. It can be applied to any scene that needs to locate and understand the sound source.

Imagine you are in a busy, noisy market. It is probably a challenge to hear what people around you are saying. With this technology, the voice of the companion can be recognized by a deformable hearing aid and then translated into words that you can understand. It can also filter out the noise of the downtown and the voices of other people.

Biosonic algorithm for bionic bat ears

The bat's ultrasound has a frequency of 10 – 200 kHz, while the human ear can only recognize sounds of 20 Hz – 20 kHz. So for us it's too loud, most of which can't be heard. In order to control the bat's biological sonar frequency and accuracy, Gupta joined the Han Jinping team to write code that converts the voice signal into ultrasonic pulses – and then transforms the ultrasound into the normal language we can hear.

This first requires the creation of a database. In order to be as simple as possible, the Han Jinping team used the pronunciation of English letters and numbers of 11 American English readers in the open source database of Carnegie Mellon University.

For example, data appearing in the form of the letter "A" or the number "1" is received by the microphone and then converted into an ultrasonic signal. The ultrasonic speaker plays the signal. Subsequently, the artificial bat ear with "dynamic outer ear" receives the signal. Finally, the software converts the ultrasonic signal into raw data—the letter “A” or the number “1”.

Artificial ear in reality (noisy city)

Although the sound signal database containing only letters and numbers has a large limitation, by analyzing it, the Han Jinping team said that the artificial ear enriches the speech signal with a "dynamic, directional time-frequency model". Next, the researchers compared the sound of the artificial ear treatment with the original voice to measure the accuracy of the artificial ear. Therefore, they put the original speech data and the sound data processed by the artificial ear into a classifier for identification. 67% of the voice signals can be successfully identified. In the control group without the dynamic outer ear, only 35% of the sound data was identified.

With more available analytical data, researchers will begin to test the system with industry benchmarks and develop bionic learning algorithms. Or, in the future, they might develop a "listening" app that turns the smartphone microphone into a directional microphone that connects to the Internet of Things to help the user choose the sound he wants to hear in reality. IBM researchers believe that implementing it is not too far away.


Car Batteries

Accumulator Products,Lead Acid Batteries,Traction Batteries,Stable Power Car Battery

Shaoxing Honyo International Trading Co., Ltd , https://www.honyopower.com