Speech quality estimation of voice over internet protocol codec using a packet loss impairment model

Area of Science:

• Speech processing
• Telecommunications
• Signal analysis

Background:

• Packet loss significantly degrades speech quality in digital communication systems.
• Packet Loss Concealment (PLC) algorithms are crucial for mitigating these impairments.
• Existing models may not fully capture the nuances of PLC-induced quality degradation.

Purpose of the Study:

• To propose a novel degradation and cognition model for estimating speech quality impairment caused by the SILK CODEC's PLC algorithm.
• To analyze the impact of packet loss on different packet classes and their relation to PLC.
• To develop a Mean Opinion Score (MOS) based estimation of total quality degradation.

Main Methods:

• Analysis of quality degradation across various previous and lost packet classes.
• Measurement of PLC effects using class conditional expectation of degradation scores.
• Derivation of a cognition module for MOS-scale quality estimation.

Main Results:

• Developed an encoder-based class model with a correlation coefficient of 0.93 with subject test results.
• Developed a decoder-based model with a correlation coefficient of 0.87 with subject test results.
• Demonstrated the model's effectiveness in predicting speech quality impairment.

Conclusions:

• The proposed model accurately estimates speech quality degradation due to PLC in the SILK CODEC.
• The encoder-based model shows higher correlation with subjective listening tests.
• This approach offers a valuable tool for optimizing speech transmission quality in packet-switched networks.