New Study Devises Enhanced Auditory Processing System for Speech Recognition


Date : Mar 01, 2019 Author : PMR Editorial Staff Category : Automation

New Study Devises Enhanced Auditory Processing System for Speech Recognition

A group of researchers has designed a system leveraging biomimetic binaural sound source localization to examine how physiognomy influences sound recognition in humans. The researchers explained that the human auditory system works in complex ways and has the ability to pinpoint the source of audio in crowded places by separating the source of sound from the background, spatially locating them, and detecting its motion before identifying the context.

The external part of the human ear helps in collecting and reflecting sound waves and modifies its frequency based on the source’s distance and location. Sound waves travel to the internal ear where they are picked up by Corti which senses the waves and turns them into nerve impulses. From there, the nerve impulses are forwarded to the medial superior olive (MSO) and lateral superior olive (LSO). While the MSO helps detect the location of the source of the sound, the LSO is used to segregate the source from the crowd. In the end, the signal is relayed to the brain’s inferior colliculus (IC).

A group of researchers has designed a system leveraging biomimetic binaural sound source localization to examine how physiognomy influences sound recognition in humans. The researchers explained that the human auditory system works in complex ways and has the ability to pinpoint the source of audio in crowded places by separating the source of sound from the background, spatially locating them, and detecting its motion before identifying the context.

The external part of the human ear helps in collecting and reflecting sound waves and modifies its frequency based on the source’s distance and location. Sound waves travel to the internal ear where they are picked up by Corti which senses the waves and turns them into nerve impulses. From there, the nerve impulses are forwarded to the medial superior olive (MSO) and lateral superior olive (LSO). While the MSO helps detect the location of the source of the sound, the LSO is used to segregate the source from the crowd. In the end, the signal is relayed to the brain’s inferior colliculus (IC).

In a bid to replicate the complex procedure using an algorithm, the researchers designed an elaborate machine learning framework that mimics the human auditory system and installed it in a humanoid robot. The complex framework was divided into four layers. The first layer was an SSL component used for decomposing the audio into a set of frequencies and generate the signals in spikes mimicking the neural impulses created by the Corti. A model imitating the functionality of MSO formed the second layer while the third layer comprised of components that worked as the LSO for the machine. The final layer of the framework was designed to mimic the IC of the brain. In addition to this, an extra neural network was kept in place to minimize the sound produced by the motion of motors operating inside the robot.

The system was tested for sound source localization (SSL) and automatic speech recognition (ASR). After testing the system thoroughly, the researchers concluded that the system works robustly and enhances the audio processing in robots and can work effectively in real-world instances without the requirement for adding additional computational power.

The efficient working of the system is a key breakthrough for the non-medical biomimetic market which is estimated to surge owing to the development of such innovative solutions. Biomimetic robots are finding increasing use in non-medical fields such as defense where they serve as an efficient medium for surveillance. Additionally, enhanced biomimetic robots are also being used in the defense sector for logistical purposes aiding soldiers in carrying heavy weaponry. Owing to these factors the defense sector is estimated to account for the share of the non-medical biomimetic robots market. However, industries such as oil & gas, water treatment, and mining are likely to bolster market growth significantly. The growth can be attributed to biomimetic robots providing an easy and efficient way of monitoring performance and leakages of pipelines utilized in the industry. Substantial investments in research and development for innovative solutions to automate the process of pipeline maintenance and leakage monitoring is further estimated to contribute to non-medical biomimetic robots market proliferation. The factors are estimated to drive non-medical biomimetic robots market growth which is likely to be valued at US$ 3 billion at the end of 2024.   


Back To Top