Phone-dependent channel compensated hidden Markov model fortelephone speech recognition

摘要

We propose the phone-dependent channel compensated hidden Markovnmodel (PDCC-HMM) for telephone speech recognition. The PDCC-HMM isnderived by modifying the conventional hidden Markov model (HMM) with thenphone-dependent channel compensation vectors. The telephone speech isnrecognized efficiently by using the derived PDCC-HMM. Experimentsndemonstrate the robustness of PDCC-HMM in speech recognition and shownthe significant reduction of recognition error rate by 50% compared tonthe conventional HMM method