Audio and Speech Processing

Bleeck S., Karbasi M., Kolossa D.

Developing automatic systems that can improve noisy speech is hard. In order to facilitate the process, it is necessary to know how well humans understand speech in noise. However, such measurements are time consuming and expensive. We have developed an automatic system to predict how well people understand speech in noise, which unlike most other methods, does not need the clean signal as reference (that is usually not available in the real world) but instead we use machine learning (based on automatic speech recognition like Siri on the iPhone) to predict how we would hear in the real world. Our method has been demonstrated to be better than commonly used systems and has the potential to be used much more often in future.

---