Published on

Speech without background noise

  • Name
    Dorothy Hardy
    COG-MHEAR Research Programme Manager

Hearing aids are signal processing systems. They take in audio signals and amplify them. The aim of the COG-MHEAR research programme is to get the parts of the audio signal that you want to hear to you, with distracting noise removed.

Dr Erfan Lowemi joined the research programme recently and started off by giving a talk about ways to make good signal processing systems. He explained that the perfect information processing system needs to pass on a signal and discard any noise. The signal is anything useful that you want to know about, such as a speech signal that contains emotion and accent as well as the actual words being spoken. The system needs to work out what is noise and what is wanted, in each different situation.

You might also want to know what is happening in the background, so that you get the context of where the speaker is, and what is going on. When recording speech, it could be useful to know if someone was speaking from a moving vehicle or a forest, for example. Then some of the background ‘noise’ is suddenly not noise: it is something that you would like to have in the signal.

A good signal processing system needs to filter what you need from what you don’t want. A filter is software (computer code). Systems that use filters are designed and trained to select what is needed and discard other information. Filters can be designed to imitate the cochlear (a part of the inner ear), so that sound is processed in a similar way to the way in which they would be in a human ear. Systems can even be set up to train themselves to work optimally. Outputs can only be as accurate as the data that is entered. The input needs to include all the information relevant to the task. A filter can be considered as a chain of transformations. At each stage, useful data can be lost. If you lose something initially then it is lost forever.

Developments in speech processing are ongoing. Systems are now being developed to understand speakers with speech impairments. This is a key step in widening communication possibilities and accessibility.