Enhancing Answering Machine Detection through Adaptive Speech Transcription: A Telecommunications Industry Solution

This study examines an innovative approach to Answering Machine Detection (AMD) leveraging the PocketSphinx speech recognition system, a lightweight and efficient tool based on Hidden Markov Models (HMM).The research conducts a comparative analysis between traditional cadence-based methods and a speech recognition system, trained on a specialized dataset of answering machine responses.The developed solution achieves high detection accuracy (90-98%) compared to traditional methods (70-80%) while maintaining optimal server resource utilization through its lightweight architecture.Key technical features include preconfigured analysis parameters, integration with Asterisk PBX through EAGI script, and support for both cloud and on-premise deployment.This research contributes to the field by demonstrating how efficient HMM implementation with specialized training data can provide high-performance AMD solutions while minimizing computational overhead.