A method is disclosed that enables the monitoring, evaluation, and adjustment of a telecommunications network's audio-signal loss plan. The method can be implemented at a data-collection server, in which the server accumulates voice-quality measurement statistics from various nodes in the network. Such nodes include telecommunications endpoints, media gateways, private-branch exchanges, teleconference bridges, and so forth. The different types of statistics that can be acquired include voice activity detection, average speech level, average noise level, and so forth. The server accumulates the statistical data from the various nodes for multiple calls and over an extended period of time. The server is also able to compare the statistics against a theoretical model that is a function of the loss plan, at least in part. For example, the comparisons that the data-collection server performs can be used to determine why certain calls have been reported as having unsatisfactory quality.