TU Ilmenau develops bio-inspired microphone for speech recognition

TU Ilmenau develops bio-inspired microphone for speech recognition
TU Ilmenau develops bio-inspired microphone for speech recognition
The Technische Universität Ilmenau, together with research partners, has developed a microphone inspired by biology that picks up sound similar to the human ear. The microphone could help improve speech recognition for controlling a variety of digital applications. In the future, the new process could even make overall acoustic systems consisting of a microphone and speech recognition more efficient, so that they consume less energy. The results of the research conducted by the Group of Microand Nanoelectronic Systems at TU Ilmenau and its research partners have just been published in the renowned international journal Nature Electronics.

Voice recognition technologies such as Alexa or Siri have made rapid progress in recent years. They can now understand voice commands very well, for example, to control cell phones, operate an alarm clock or even a wide variety of smart home applications. However, noisy environments with many different sounds, such as train stations, restaurants or streets, are still very problematic for current technologies with conventional microphones. The human ear, on the other hand, is able to pick out the voices of individuals in noisy environments with a lot of background noise and understand what is being said.

A team of researchers from the TU Ilmenau, the Christian Albrechts University of Kiel, the Karlsruhe Institute of Technology, the University College Cork and the Ilmenau Fraunhofer Institute for Digital Media Technology has now made use of the advantages of the human ear. In the process, the scientists are replicating the function of the inner ear, which supports hearing comprehension in humans, especially in noisy environments. The bio-inspired microphone they developed uses three properties of the human auditory system:

Sounds, such as a spoken word or a melody, consist of different tones. They can be recognized by speech recognition systems by decomposing the sounds into the tones they contain. In humans, this decomposition takes place not only in the brain, but already in the inner ear. This consists of a large number of small hair cells, each of which reacts to different sounds. This prevents the different sounds from influencing each other during hearing and prevents soft sounds from being masked by loud sounds. Unlike conventional microphones, the bio-inspired microphone breaks down sounds before converting them into an electrical signal.

in order to understand soft sounds, they must be amplified significantly. Loud sounds, on the other hand, in order to be perceived undistorted, must be attenuated. This ability of the auditory system allows humans to hear sounds whose sound pressure varies by a factor of one million. In the inner ear, this what experts call nonlinear or compressive amplification is already integrated into the sensor. This makes it possible to hear significantly wider ranges of loudness.

In quiet environments, it is relatively easy for people to pick out sound signals that are important to them, such as the voice of a single person. In noisy environments, however, this is difficult and the sound signals that are important to the listener must be amplified more than the ambient sounds for good listening comprehension. Therefore, in order to hear as well as possible in different environments, the sensors of the inner ear, the hair cells, are permanently adapted to the changing requirements. This is achieved by changing the relative amplification factors for individual sounds depending on the situation.

The technological heart of the bio-inspired microphone is silicon bending beams modeled on hair cells, ranging in length from one-third of a millimeter to just over one millimeter. This difference in length causes each bending bar to respond to only a single tone of the sound signal (point 1). An electronic control system makes it possible to control the characteristics of each bending beam, such as amplification, separately (point 2) and thus to adapt them to different environments, for example those of different loudness (point 3).

Thanks to the automatic adaptation to the sound signal, which can be controlled separately for each pitch, the relevant signals are highlighted. As a result, according to the researchers’ idea, less computing power is required for speech analysis and thus less electrical energy for the overall system consisting of speech recognition and microphone. In particular, applications with limited energy capacities, such as hearing aids, would thus be able to perform significantly more complex speech and sound analyses. In addition, the production of the bio-inspired microphone based on silicon technology not only enables micrometer-precise manufacturing, but also low-cost mass production.

Together with scientists from partner institutions, the TU Ilmenau now intends to develop prototypes of the bio-inspired microphone. Possible applications include speech assistance systems and hearing aids as well as technologies for monitoring machines in production.

The demonstrator of the new bio-inspired microphone was realized in the "ForLab Ilmenau for Neuromorphic Electronics", which is supported by the Federal Ministry of Education and Research. The ForLab is affiliated with the Center for Microand Nanotechnologies (ZMN ) at TU Ilmenau.

The research results published in Nature Electronics were generated as part of two major research projects at TU Ilmenau: " SFB 1461 - Neuroelectronics: Biologically Inspired Information Processing," which is funded by the German Research Foundation, and "MemWerk ," funded by the Carl Zeiss Foundation, in which TU Ilmenau is researching smart materials for biologically inspired electronics.

Publication in Nature Electronics:

’023 -00957-5

DOI number: 10.1038/s41928’023 -00957-5