Waveform based Voice/audio recognition system on MATLAB software and LED based hardware
Video explanation of practical working model of voice recognition system
This program is a collection of programs in the MATLAB software. It has a main program which calls all sub programs that constitute a software which is MATLAB dependent.
The main program has been named- Detect_Voice Some arguments are passed from main program to define input to output in which are parameters are:
- Recording time (in seconds)- How much time the input voice signal is recorded.
- Waveform width- How much are of the captured waveform will be visible.
- Waveform moving rate- The rate of movement of waveform
- Local threshold- Noise removal tool. An audio signal consists of 2 types of noise: high frequency and low frequency noise.To remove both these noise signals we can use from a variety of options which are:
- Discrete Fourier transform
- Discrete cosine transform
- Discrete wavelet transform
- And many more
Here in this video we have used Discrete Fourier transform to remove the low and high frequency signals so that we only have our main signal left.
We have here with us 4 option i.e. 4 possible outputs that we will shown on the screen and the LED based hardware. The 4 possible outputs are A, AB, ABC and ABCD.
To start the program first we define a port to make a connection between the MATLAB software and the hardware system. The port is activated and connection is made which switches on the green LED. Now, we have to press any button to start the recording. The audio is given through a common 3.5mm jack earphone. When the program is completed with the recording it gives us the option to either see the waveform and then see the output or directly see the output. In the video by pressing 0 we proceed directly to see the output.
The output appears on the laptop screen in form of that alphabet as well as on the USB port connected hardware device. The function of the hardware device has been simulated in the MATLAB software and both are capable of working independently. The 4 LED's glow corresponding to the output result.
The important point to note in this project is that the program is not exactly speech recognition but just alphabet recognition where the software and the hardware identifies the number of alphabets spoken and not the exact alphabet. It does so but analysing the input waveform of out audio input.
Functioning of this voice recognition system
The audio signal is in analog form which must be converted into digital form so that it can be analysed by the software and hardware. For this we do the sampling of the audio signal. The sampling rate set by us is 10,000 samples per second. Each sampled data is stored in a 16 bits memory cell. The memory taken in completely out of choice and can be 8 bits or 32 bits also.