Implement The Gaussian Mixture Model

Published Date: 02 Nov 2017

The overall working of the GMM model can be understood by the above figure. To implement the model we have take some model. Then we will train this model on the bases of the some emotions. We take three emotions and every emotion consists of the ten samples. The first aim is to collect the samples and saved them. Then the second step is to implement the model and third step is to execute the model. To implement the GMM we have to take write some code in C# Language and execute on the MATLAB tool. There are some specific codes that can be writing on the interface of the panel zone of the MATLAB. These codes are written for different basic models. Each coding has a set for different model.

Step3: Test for Unknown Speech Samples

To test the unknown speech, first we take a sample for audio speech. This sample may store in the form of digital form. Then we collect it with a location in database. Then we will use the wavesurfer to edit and pick the range (Time Limit). These time intervals we decide about the nature and class of emotions. Then we put this sample to GMM this model will display a confusion matrix. In this confusion matrix we can check the higher number of occurrence with the numeric values. By this confusion matrix we can classify about the unknown speech. This is the one method to identify this. The other attributes are also support to identify the speech. Speech has already its own features.

Step4: Transfer of Emotional Signal

The actual work has start from this phase. This phase basically work like an interface in between the machine and human. Human want to control the activities of a machine. After saving the emotions in database the next task for machine comes to match and display the activities for the emotional state. These emotional states are the generic reason for the machine controlling. The Interesting thing is that nobody and no machine can set every situation at all time. The basic phenomenon is to set such kind of environment in front of the machine.

Step5: Control the Machine through Emotions

Now for controlling the machine we have to set some emotional environment for the machine. To handle this situation we have two choices. This can be done by the help of man as well as a machine. It may be possible for a machine; it can become dangerous for that time period. So it will be risky for human to come up in the front of the machine.

As on the second hand if a person can go in front of the machine. This will be depending on the person how much capability he has to be control the emotions. Because till now, the result shows that except technical reason, machine works better as compare to person. There are two main reasons behind the working of a machine. These are the speed and accuracy.

To control on these parameters man is working on it continuously. Almost maximum part of for controlling the machine by emotions is depending on the Emotional Database (ED) and Learning concept of the machine. The one aspect to behind develop and produce a machine is that we are not building any Negative kind of criminal activities for the machine. To control the machine we have so many practical implementation approaches. If we are able to collect a huge and robust database that has all kinds of criminal issues and possible activities related to crime, Blackmail, bed behavior, miss behavior and any destructive job, then it will be very beneficial for controlling machine. Because this is the fact that in any emotional state a man can be controlled in that emotions state only. Some of them are as following.

3.5.4.4 Emotion Control System (ECS)

The proposed approach is based on emotions. A machine can be control by passing emotions to the machine. The main idea behind this learning is to develop such kind of system. This can be implemented on the development of Emotion Control System (ECS). In this methodology there will be an electronic chip that will build

for controlling the emotions. This chip or device will be called as Emotional Chip (EC). This electronic chip can be embedded for some basic emotions. After testing this trial demo, this can be extending for further aspect. This can be done by taking with a small experience on the building a small electronic chip.

In the starting state the EC will be inserted in the machine and stay in disable mode. The machine cannot enable this chip by itself. This will be activating only and only from other person or other machine. All the rights and power will be assigned to server side. Mostly the preference will be awarded to human. There will be a server that will control this particular chip. For the communication to that machine the server will first send the activation message. After accept this activation message now machine can be control. This whole process will work under the wireless network. To establish the communication process the range or signal for the network is mandatory. There should be May kind of mechanism that can handle the

Communication process. There will be some advantageous and disadvantageous that governs the communication process. In this way man, machine and Network are to have some use of the electronic power.

Collect the phrases.

Collect the religion literature.

Collect the emotional pictures and images.

Construct the dressing gadgets.

Make some building and room environment.

3.6 Discussion

This section will describe all the main parts and causes that are implementing above. The basic requirement for the machine learning is to implement such kind of mechanism that can act as like a human. Till now Japan is for upcoming country that is making this kind of robots. Japan has invented a robot name ASIMO. This robot is look like a human. To handle the machine (Robotic machine) the server will send the activation message. Then machine will open its port for communication to other device. The effects will be discussing letter on.

This chapter basically illustrate about the analysis and results. This include of the chief experimental work resulting from the experimental design that is deeply describe in the previous chapter. The next heading from 4.3 to 4.6. The training of the model and test of the model are major in the emotion system. The first heading may have several parts like collection of data. The collection of data is may of different kinds. The emotional data base should be effective for man as well as machine. In the first task the collection of the different kinds of emotion are prescribed in various forms. This data we can say that dirty data in our case. Dirty data for our model is apart from the three emotions (Happy, Anger, and Sad). The data can have any emotions apart from the three emotions that data is consider under the dirty data. So our first aim is try to keep such data only.

To collect the data we can also take the help of any software tool. These tools are basically helpful in collecting the data. These tools are generally open source software. These tools are freely available on internet. Some of them are like wavesurfer, praat etc. The thing very important in case of collecting data is that data should be homogeneous. Our collecting data may have of different types such as .wave, .avi, .mp3, .mp4 etc. The first stage decides how much perfect our system will run. The model is trained only for that data.

4.2 Overview

Hence the result part shows the concept of machine learning. This approach is a basic fundamental towards machine become emotional. The emotions are very specific features for anyone. Maximum things are there that can be control by emotions. This is the main aspect of this approach. Japan is leading country in this field. The engineers from Honda Company are created a robot name ASIMO with 34 Degree of freedom. This degree helps this into walk and performs like a human. It has body structure and features as following [28].

Height 130cm

Weight 54 Kg

Walking speed 2.7 Km/hr

Running speed 6Km/hr

Grasping force 0.5 Kg/hand

Sensors (Foot) 6 Axis Foot Area Sensors

Power Rechargeable 51.8 V Lithium Ion Battery

Operating Time 1 hr

This has a wide scope in future. The achievement of the today’s working of robots. This all depends on today’s progress in this field. Science gives the new concept and engineering is the platform that implements these concepts. Science is the only conceptual object just on the piece of paper. But the engineering is the field that has the ability to implement these concepts. Engineering is the domain that builds something that never happened before. Not only implementations of the concept but also relate to the real life. In real life we don’t have so much energy to make this system. These systems are made once and work for a long times.

In the present approach we discussed about the learning of a machine from emotions. A machine can learn from emotions. But the major thing is that how a machine can learn from emotions. What are assumption and way by which emotion can generate in machine? Robot or any machine that has some power system. This power can be manage by power management system. This system has some policies and some strategies how to save more power. The emotions are very sensitive and specific for anyone. These are valuable facts. These are not general tasks. A human and machine both can have some creative ideas. How to implement these ideas into the beneficial for human life. There should be no hesitation should be for machine.

4.3 Collection of sampling data

The first step for this model is the collection of the data. This data has several specific features. The general formats are given above. In our case our model is trained on the waveform on mono waveform. This step is partially describes in the above for the working of wavesurfer in section 3.5.2. We generally take the sample of length for 0.20 ms to 01 or 02 seconds. We are able to remove the gap between two and more clip sample for the emotions. Generally we avoid the gape or silence speech of the sample. The sound wave should be in mono wave form.

So in this step we take a long clip (02-05 ms) for clipping. Then by the help of software tool we open this file in file browser. After open we see the pictorial waveform for the sound. As per need after listening the sound, we cut the required amount of the clip. Then check the format for the cutting clip and save this clip as a wave file in destination folder. Save all the cut files in the same file. The number of files for each emotion should be same.

4.4 Features Extraction

Once we have created the emotional database, the feature extraction phase comes. This phase/step consider the maximum extracting features for the saved files. Among the huge amount of data, how and which sample should be taken. Features extraction is also a part of the collecting the data. This phase collect the sample of files those contains almost same kind s of the features. This will make the model that can works on this kinds of files. All the properties of the sample files are homogeneous. The performance of the model is based on the input data. If the data is homogeneous then it provides maximum likelihood expectation.

4.5 Training the model

As the name suggest that there should be training for the model. The development of the model is done under last two steps. Now our model should be getting trained. For training the model there is a database that contains some rules and properties. The main objective is to make some kind of mechanism that can fire some action when some properties matched to the properties. This training can be improved further by the help of the learning process. The after all main agenda is somehow to make the machine that can easily understood and recognize the emotions for the unknown speech sample with some emotions.

4.6 Test the model

Testing is very essential to the any model or system. When system developed for the objective then testing play an important role. Testing part describe how our system is working and how much it giving the desired result. Testing may be of several kinds. Some of the types of the testing are unit, integrated, utility based, white box, black box and system testing. Basically testing part allows the error free and desired output from the unknown speech. In this process/step we provide some input files as unknown files the after executing and compiling the coded to the data it gives us the confusion matrix. This confusion matrix shows us how much our input files are matched.