Coded Speech Quality Measurement by a Non-Intrusive PESQ-DNN

Wideband codecs such as AMR-WB or EVS are widely used in (mobile) speech communication. Evaluation of coded quality is often performed subjectively by an absolute category rating (ACR) listening test. However, the ACR test impractical for online monitoring communication networks. Perceptual evaluation (PESQ) one metrics instrumentally predicting results PESQ algorithm requires original reference signal, which usually unavailable network monitoring, thus limiting its applicability. NISQA a new non-intrusive neural-network-based measure, focusing on super-wideband signals. In this work, however, we aim at well-known metric using xmlns:xlink="http://www.w3.org/1999/xlink">PESQ-DNN model. We illustrate potential model scores wideband-coded obtained from operating different bitrates noisy, tandeming, and error-prone transmission conditions. compare our methods with state-of-the-art topologies xmlns:xlink="http://www.w3.org/1999/xlink">QualityNet , xmlns:xlink="http://www.w3.org/1999/xlink">WaweNet xmlns:xlink="http://www.w3.org/1999/xlink">DNSMOS —all applied to prediction—by measuring mean error (MAE) linear correlation coefficient (LCC). The proposed offers best total MAE LCC 0.11 0.92, respectively, conditions without frame loss, still when including loss. Note that could be similarly non-intrusively predict POLQA other (intrusive) metrics. definition code provided https://github.com/ifnspaml/PESQDNN .