SPEECH RECOGNITION DEVICE AND SPEECH RECOGNITION METHOD

摘要

There are provided a speech recognition device and a speech recognition method capable of improving the noise processing function in speech recognition and reducing the memory amount used. In advance, an acoustic model is subjected to clustering processing, and a center of gravity of each cluster and a difference vector between the center of gravity and each model are calculated. Estimated noise models are model-combined with the calculated center of gravity and the combined center of gravity and the difference vector are stored in a memory. In actual recognition, an optimal center of gravity for the environment estimated by an utterance environment estimation is extracted from the memory. By using the difference vector stored in the memory on the center of gravity extracted, a model restoration is performed and noise processing is executed by using the model.