Long Short-Term Memory Recurrent Neural Network for Automatic Speech Recognition

Abstract

Automatic speech recognition (ASR) is one of the utmost demanding tasks in Natural Language Processing due to its complexity. Recently, deep learning approaches have been deployed for this task, and have been proven to outperform traditional machine learning approaches such as ANN. Particularly; deep learning methods such as Long Short-Term Memory (LSTM) has achieved improved performance in ASR. However, this method is limited in processing continuous input streams. Traditional LSTM requires 4 linear layers (MLP layer) per cell, which require large amounts of memory bandwidth to run at and for each sequence time-step. LSTM cannot afford many computational units required in processing continuous input streams because the system does not have enough memory bandwidth to feed the computational units. In this research, an enhanced deep learning LSTM RNN model is proposed to resolve this shortcoming. In the proposed model, a Recurrent Neural Network (RNN) is incorporated as a "forget gate" to the memory block to allow resetting of the cell states at the beginning of sub-sequences. This will enable the system to efficiently process continuous input streams without necessarily increasing the required bandwidths. In the proposed model, the standard architecture of the LSTM networks has been modified to make effective use of the model parameters to address the computational efficiency problems of large networks on large vocabulary speech recognition. Some CNN based models and Sequential models were also used on the same dataset, and the performances of the models were compared with the performance of the proposed model. The LSTM-RNN outperformed the other deep learning models with the accuracy of 99.36% on the well-established public benchmark spoken English digits dataset.

Performance Comparison of Various Neural Networks for Speech Recognition

“…The performance of Automatic voice recognition is primarily affected by pooling size rather than pooling layer overlap [2].The CNNbased strategy for voice recognition outperforms the traditional ANN-based approach in terms of accuracy [3].Based on features extracted from the Bark spectrogram, a convolutional neural network (CNN) is utilized to categorize spoken digit recognition data [4].By creating phone classes from raw speech signals, CNN performs better than other parameter-based approaches [3]. A unique technique called the stride-based convolutional neural network (SCNN) reduces the number of convolutional layers and eliminates the pooling layers in order to boost computational stability [5].Using knowledge distillation, it may be possible to convert a BiLSTM model into a low-latency end-to-end UniLSTM model [6].Li-GRU layer significantly lowers computational complexity and boosts recognition performance, saving more than 30% of training time compared to a standard GRU [7].There is a trade off between learning rate and accuracy when trained through LSTM-RNN [8].Using a 15-layer deep network, convolutional LSTMs obtain a word mistake rate of 10.5% without using a dictionary or language model [3].…”

“…With the help of this property, LSTMs may process whole times series data sequences without considering each data point separately. Instead, they can process a new data points by using the information from past data in the sequence [23].The previous state of the input sequence is kept in the memory cells of the LSTM.…”

The Application of Cloud Computing Service and Voicematic Identification Intelligence in the Supply Chain during the COVID-19 Pandemic

“…ey should enhance their design capabilities, create their intellectual property rights, and shorten product lead time and new product launch cycle to reduce customers' costs and improve their sales efficiency. [19].…”

Supervised Attention-Based Bidirectional Long Short-Term Memory Network for Nonlinear Dynamic Soft Sensor Application

“…It has been verified that the LSTM can better deal with gradient disappearance and explosion caused by long-term dependence. 27 As a result, the LSTM has gained widespread interest in natural language processing, 28 speech recognition, 29 and industrial process time series data. 30 Since the input of the LSTM focuses only on the dynamic information on process variables, the introduction of quality information into the cell structure to enhance the dynamic extraction of quality information is not considered, which makes the correlation between the hidden state and dynamic quality information on the LSTM network biased.…”