Speech Recognition Using HMM-CNN

Nowadays, the largest part of speech recognition systems is based on statistical modeling, being Hidden Markov Model (HMM) the most popular among them, followed by Convolutional Neural Networks (CNN), whose learning is based directly on the raw signal. This work presents a study comparing HMM and CNN approaches and a hybrid approach based on HMM and CNN for speech recognition under noise, comparing metrics, such as accuracy, response time, and computational cost to generate models. The experimental results show that the integration between HMM and CNN increased the accuracy by 6% and 8% when compared to HMM and CNN isolated, respectively.