Real-Time Speech Enhancement Based on Convolutional Recurrent Neural Network

Speech enhancement is the task of taking a noisy speech input and producing an enhanced output. In recent years, need for has been increased due to challenges that occurred in various applications such as hearing aids, Automatic Recognition (ASR), mobile communication systems. Most Enhancement research work carried out English, Chinese, other European languages. Only few works involve Indian regional Languages. this paper, we propose two-fold architecture perform Tamil signal based on convolutional recurrent neural network (CRN) addresses real-time single channel or track sound created by speaker. first stage mask long short-term memory (LSTM) used noise suppression along with loss function second stage, Convolutional Encoder-Decoder (CED) restoration. The proposed model evaluated speaker environments like Babble noise, car white Gaussian noise. CRN improves quality 0.1 points when compared LSTM base also requires fewer parameters training. performance outstanding even low Signal Noise Ratio (SNR).