Review of Literature on Controlling Device for Smart Homes Using Speech Recognition Through Mobile Phone

Introduction

In our day to day life there are several appliances which are controlled by remote controlled devices. There is problem for elder and handicapped person to recognize the remotes for different devices as most of them are same in shape and size. If take an example of visual impaired people first it is difficult for them to recognize which remote control belongs to which home appliances and after that to recognize the target device which is assembled along with the home appliances. After reviewing many papers I found that a solution for this problem can be obtained with the devices which are controlled with the help of gesture and voice recognition. One solution that comes in everyone’s mind that is Smart Home. Yes it is a great concept to be implemented but can you imagine what would be the cost of this? It is a great innovation but it will be somehow restricted to The Elite Group. So if want a real transformation we have to make it available and affordable for every economic group.

There are many possible solutions for this problem. But still these types of homes are not very popular? The reason is very simple cost! Instead of making a new product we can make it much cheaper by using existing resources. Nowadays mobile are the common gadgets that we have in our home. Can make a good use of this resource as it already comes along with a microphone.

Literature Survey

Personalized Speech Recognition On Mobile Devices, Ian McGraw, Rohit Prabhavalkar, Raziel Alvarez, Montse Gonzalez Arenas, Kanishka Rao, David Rybach, Ouais Alsharif, Hasim Sak, Alexander Gruenstein, Francoise Beaufays, Carolina Parada.

In this paper authors have done research work on making such a recognition system that is accurate, low latency with small memory along with a computational footprint which will help in run faster on android devices. Quantization has been done on Long Short Term Memory (LSTM) with Connectionist Temporal Classification (CTC) which can directly analyze and predict the phoneme targets. Here its memory size has been reduced by SVD based compression scheme. Here the base concept is quantized deep neural networks (DNNs) and on-the-fly language model rescoring to achieve real-time performance on modern smart phones. In the article small size of memory and computation constraints, the result of word error rate (WER) performance and latency by employing Long Short-Term Memory (LSTM) recurrent neural networks (RNNs), trained with connectionist temporal classification (CTC) and state-level minimum Bayes risk techniques is very appreciable and highly accurate. LSTMs are made small and fast enough by quantizing parameters to 8 bits, by using context independent (CI) phone outputs instead of more numerous context dependent (CD) phone outputs, and by using Singular Value Decomposition (SVD) compression.

Acoustic Models are trained on 3M hand-transcribed anonymized utterances extracted from Google voice search traffic (approximately 2, 000 hours). All models are trained using distributed asynchronous stochastic gradient descent (ASGD). To improve robustness to noise, they generated “multi-style” training data by distortion of each training utterance using a room simulator with a virtual noise source, for generating 20 distorted versions of each utterance. They extracted from YouTube videos and environmental recordings of daily events for Noise samples.

For memory consumption reduction further, they compressed acoustic models using projection layers that sit between the outputs of an LSTM layer and both the recurrent and non-recurrent inputs to same and subsequent layers. By adapting the acoustic Models to generate multi-style training as described above results in a further 12. 8% relative improvement over the SVD compression model.

Since the 11. 9 MB floating point neural network acoustic model consumes a significant chunk of the memory and processing-time, After Quantization of model parameters into 8-bit integer-based representation had an immediate impact on the memory usage, reducing the acoustic model’s footprint to a fourth of the original size. The final footprint of Acoustic Model was 3 MB. For on device language modelling focus is on building a compact language model for dictation and voice commands. Maintaining a small system footprint, they trained a single model for both domains. They also limited the vocabulary size to 64K. Language models are trained using unsupervised speech logs from the dictation domain (∼100M utterances) and voice commands domain (∼2M utterances). This design of a compact large vocabulary speech recognition system can run efficiently on mobile devices, accurately and with low latency. This was done by using a CTC-based LSTM acoustic model which predicts context-independent phones and is compressed to a tenth of its original size using a combination of SVD-based compression and quantization. For efficient decoding, we use a on-the-fly rescoring strategy following with additional optimizations for CTC models which reduce computation and memory usage. The combination of these techniques allows to build a system which runs 7× faster than real-time on a Nexus 5, with a total system footprint of 20. 3 MB.

Remote Control System of Home Electrical Appliances Using Speech Recognition, Noriyuki Kawarazaki and Tadashi Yoshidome.

In this paper mainly developed remote control system of electrical devices with the help of speech recognition. Remote control system is consisting of the PMRC, a PC, a microphone and a speaker. PMRC is programmable multi remote controller which is used here can memorize the functions of many remote controllers. PMRC is a device which can perform the task of several remotes at once. It has infrared LED’s mounted all around it. These infrared LED’s sent infrared signals in any direction. So user doesn’t need to worry about the position of the PMRC.

When a user gives the speech command to the system, the PMRC sends the infrared ray’s signal to the electric home appliance. Then system gives the voice synthesis message to the operator and many speech commands for the remote control operations depends on sentences so as to have a human friendly interface in the system. In this use the voice recognition software and a morphological analysis software in order to recognize the speech commands based on the sentence. “Julius” is famous speech recognition free software for researchers. “Mecab” segments Japanese sentences into morpheme sequences and analyses according to part of speech.

The average rate of speech recognition is 60%. The error caused due to not understanding of commands based on sentences. In the future do the many tests for visual-impaired and old people and also define several kinds of speech commands for the practical application of the system.

Homes Appliances Controlled Using Speech Recognition in Wireless Network Environment, Mardiana B, Hazura H. , Fauziyah S, Zahariah M. , Hanim A. R, Noor Shahida M. K, International Conference on Computer Technology and Development, 2009.

In paper Proposed system comprises of client system, server system and wireless interface. Interface allows client system to take user’s speech as an input and then transmission of that captured voice command to voice recognition server. Interface allows client system to take user’s speech as an input and then transmission of that captured voice command to voice recognition server then Recognition server translates voice into symbolic data file which is send to both the sender user as well as to the interface circuit via parallel portsUser has to speak in the same fashion as it can be detected by voice recognition server.

This system recognizes the input commands very well and manages to give a good response if the wireless coverage have strong signal in order to get a good interfacing result but range of wireless network can also vary the performance of the system.

Voice Recognition Based Wireless Home Automation System, Humaid AlShu’eili, Gourab Sen Gupta, Subhas Mukhopadhyay, 4th International Conference on Mechatronics (ICOM), 2011.

In this paper, to develop Wireless Home Automation System using Voice Recognition. The main purpose of the system is to provide assistance to elders and disabled people. The electric devices usually at offices or at houses can be linked and controlled by the developed prototype. The technology called ASR is used for implementation by the system in association with Microsoft speech API. The most well-known system for home automation available in the market is Home Automation Living (HAL). The power of an existing PC is tapped by the software to control the home. Speech command interface is provided by it. One of the main advantages of this system is that with help of existing highway of walls in the walls, the system can send commands anywhere in the house. As most of the products sold in the market are expensive and also require a significant home make over so as to make the home automated whereas HAL is easy and inexpensive to install.

Implementation of wireless network of shortens the power consumption and enhancing the efficiency by taking assistance from Zigbee radio frequency (RF) modules is done by the system. In this use the 115200 bits/s baud rate which is the maximum composition baud rate which is provided by zigbee. in this use the vb. net application for speech recognition and use the Microsoft speech API which very efficient for speech recognition. There was total of 35 different speech command both male and female were taken as sample test command so the tests conducted over 1225 commands have 79. 8% accuracy. Whatever the system was unable to recognize properly it ignores the command and sends no signal to the device controller module.

In this paper the main challenge is efficiency and low power consumption which is achieve by the Zigbee RF module. With the help of differential pulse code modulation compression algorithm implemented the multimedia streaming via network, differential consumption pulse code modulation compression algorithm allows the flatten the speech data to half of its actual data size. The future work can be integration of GSM or mobile server to operate from the distance. In this we can add the confirmation command to the speech recognition system.

Voice Recognition Based Wireless Room Automation System, Avishek Paul, Madhurima Panja, Monalisa Bagc, Nairit Das, Rudrabrata Mitra Mazumder, Soumyarshi Ghosh, 2016 International Conference on Intelligent Control Power and Instrumentation (ICICPI) IEEE.

In this paper the main objective is Design and implement a voice recognition wireless based room automation system. The voice recognition based room automation system uses HM2007L IC to identify the voice speech or commands of a specific user. After that micro-controller sends signals to the encoder for processing.

HM2007 is a CMOS voice recognition integrated circuit that supports voice analysis, recognition process and system control functions. It is a 40 isolated-word voice recognition system that has external microphone, keyboard, 8K X 8 Static RAM. The speech recognition system is trained with the words, the user want the circuit to recognize. The system can accept voice samples from multiple users, in any language and can identify the user as well as the command by matching it with the trained voice samples.