We develop signal processing methods that ensure better speech intelligibility and less listening effort in a wide variety of applications. The technical transmission of speech is often compromised by superimposed reverberation and background noise, for example in the context of railway station announcements, mobile telephony or in-car infotainment systems. In media productions and broadcasting, deciding whether certain parts of speech are sufficiently intelligible for listeners is often a subjective matter. Fraunhofer IDMT’s software solutions analyse speech and sound according to objective measurement criteria and in real time – even in environments with variable acoustics.
Through our expertise in the field of signal processing and audio system technology, the quality of speech intelligibility can be measured and optimized. This allows media libraries, streaming services or providers of communication services to offer their customers added value, such as alternative soundtracks with better intelligibility or personalized solutions in end devices. Methods based on machine learning are used to identify audio signals containing speech, measure them in terms of their intelligibility and, if necessary, process them using our source separation algorithms. In this way, for example, dialogue in a media production can be accentuated even when background acoustics are complex and contain music or sound effects. In conference or telephone applications, speech signals can be automatically adjusted to ambient noise by means of adaptive signal processing. By taking into consideration recent findings from hearing research, with our solutions we can achieve an individually optimized sound experience for people with and without hearing impairments.