Hearing loss is a swiftly growing field of scientific research, as the amount of baby boomers who deal with hearing continues to rise as they grow older.
Researchers study people’s potential to recognize speech to better understand how hearing loss affects them. When there is reverberation, some hearing disability, or powerful background uproars, like noise or multiple speakers, it is more challenging for people in recognizing human speech.
Therefore, hearing aid algorithms are frequently utilized for enhancing human speech recognition. For evaluating such algorithms, researchers conduct experiments that aim to ascertain the signal-to-noise ratio at which a certain number of words (typically 50%) are recognized. However, these tests are tedious and expensive.
Researchers from Germany investigate a human speech recognition model based on ML (Machine Learning) and deep neural networks, published by the Acoustical Society of America via AIP Publishing in The Journal of the Acoustical Society of America.
Carl Von Ossietzky University author Jana Roßbach stated that the uniqueness of our model is that it offers a good prognosis for hearing-impaired audiences for types of noise with very different complexity and displays both low errors and high interrelation with the measured data.
Utilizing automatic speech recognition (ASR), the researchers computed the number of words per sentence a listener comprehends. The majority of people are acquainted with ASR via speech recognition tools such as Alexa and Siri.
The study included eight normal-hearing and twenty hearing-impaired listeners who were subjected to a mixture of complex noises that masked the speech. Hearing-impaired listeners were divided into three groups based on their level of age-related hearing loss.
The model permitted the researchers in predicting the human speech recognition performance of listeners with hearing disabilities with varying degrees of hearing loss for a collection of noise maskers with growing complexity in temporal modulation and resemblances to real speech. A person’s possible hearing loss could be contemplated on an individual basis.
Roßbach explained that they were pleasantly surprised that the forecasts performed well for all types of noise. When utilizing a single competing talker, they anticipated the model to have issues. However, nothing of that sort happened.
Predictions for single-ear hearing were generated by the model. In the future, the researchers will be developing a binaural model because understanding is influenced by two-ear hearing
Since these topics are so closely related, the model could potentially be used to forecast listening effort or speech quality in addition to speech intelligibility.