This paper proposes a speech recognition technique based on acoustic models which can take unknown variation factors (speaker voice characteristic, noise environment) into account at the time of recognition. Context-dependent acoustic models, which are typically triphone HMMs, are often used in continuous speech recognition systems. These methods enhance the accuracy of acoustic models via the respective modeling of phonemes according to the factors in acoustic variation. This work hypothesizes that the speaker voice characteristics that humans can perceive by listening and the noise environments are also factors of acoustic variation in construction of acoustic models, and a tree-based clustering technique is also applied to speaker voice characteristics and noise environments to construct proposed acoustic models. In speech recognition using triphone models, the neighboring phonetic context is given from the linguistic-phonetic knowledge in advance; in contrast, the variation factors as voice characteristics and noise environments of input speech are unknown in recognition using proposed acoustic models. This paper proposes a method of recognizing speech even under conditions where the variation factors of the input speech are unknown. The result of a gender-dependent speech recognition experiment shows that the proposed method achieves higher recognition performance in comparison to conventional methods.
展开▼
