Answering machine detection (AMD) is critical to contact centers utilizing automated dialer systems because most calls placed often result in pre-recorded audio from a machine or other automated system. Every audio recording or other automated system that is incorrectly detected as a live speaker may be routed to an agent for handling. As a result, agents may begin to assume an audio recording is at the other end of the call and mistakenly hang up on a live person, sound surprised, lose their train of thought, etc. AMD employs various signal processing algorithms to classify the entity that picks up a communication into categories, for example, such as answering machines, or recorded audio, and live speakers. The accuracy of these algorithms may depend on various parameters and requires trading off AMD rate and live speaker detection (LSD) rate. For example, biasing an autodialer towards a high AMD rate may result in more live speakers being classified incorrectly as recorded audio and hungup on by the autodialer and vice-versa. Some countries as well as applications, such as high-value dialing for example, do not allow or utilize AMD because of false positives. In another example, an autodialer operation may contact the same phone number multiple times in a day to try and reach a live speaker. By learning the fingerprint of a specific audio recording, the autodialer may prevent an audio recording from being repeatedly routed to an agent. If an audio recording associated with a contact is altered, however, the system may have to relearn the fingerprint of the audio recording the next time that number is dialed. To overcome this exemplary problem, the inventors have developed a system and method for learning call analysis. An example / embodiment of this system is one that is used for routing communications in a communication system, wherein the telecommunication system comprises at least an automated dialer and a telephony service module operatively coupled over a
