Speaker bias in speech quality evaluation over end-to end communication links

摘要

The speech quality is one of the most important concerns of the communication quality and reliability. The time consuming subjective methods for speech quality evaluation are replaced by the objective measurement algorithms implemented in the real-time field measurement tools. However, regardless of measurement type, the speaker bias tends to affect the assessment. The talker bias becomes annoying and undesirable in local geographical areas, especially close to the edge between "good" and "bad" speech quality. The paper analyzes different methods for compensating the speaker bias phenomenon in real time field measurement tools that are configured to evaluate speech quality within a local geographical area. An optimum method, called statistical compensation, is driven and the results of its effectiveness are presented.