Deep Learning Audio Super Resolution and Noise Cancellation System for Low Sampling Rate Noise Environment
Che-Wen Chen, Wei‐Chun Wang, Yang‐Yen Ou, Jhing-Fa Wang · 2022
The fire department's disaster relief center receives an average of hundreds of thousands of ambulance calls every year, among which the ambulance center dispatches and guides relevant ambulances and ambulance personnel to the scene through a wireless notification system. Ambulance personnel must continue to guide and communicate through radio during the rescue process in the fire scene. In an emergency, the clarity of radio listening information is the first requirement to strengthen the judgment of the ambulance personnel and strive to save lives in a short time. Usually the sampling rate of the general radio is 8KHz, and the disaster relief site is easy to receive a lot of noise, so it is very important to improve the sampling frequency of radio communication and noise cancellation. By integrating two deep learning models Unet+AFiLM [1] and I-DTLN into audio super-resolution and noise cancellation system, which can effectively remove noise in radio communication and increase the audio sampling rate to enhance the effect of speech recognition.