Basic Operation using DSP Processor

 We used the hardware kit for the first time. The kit used was TMS320F28375. Instructions for arithmetic, Logical and Shift operations were performed. The changes in the register values before and after execution were observed and noted. Also a program was voice modulation was demonstrated.

IEEE PAPER REVIEW

        AUTOMATIC WORD RECOGNITION IN CARS

           Author Name: Chafic E. Mokbel and Gerard F. A. Chollet, Member, IEEE

The widespread use of mobile technology motivated the development of robust recognition systems for car.

Speech produced and captured is contaminated by noise. A microphone is placed 50cm away from the mouth of the speaker and it capures a mixture of speech and noise.

A number of spectral representations and their associated distance measures were discussed.

The possibility of increasing the SNR is also considered. Using several microphones helps to achieve a higher SNR. But since only one microphone is supposed to be available for the work reported,

spectral subtraction was an appropriate technique.

A number of techniques in the domain of speech recognition in cars were verified. One important was related to Lombard effect which depends on the mean frequency characteristics of the ambient noise.

PATENT REVIEW

                 Spoken word controlled automatic dialer

PATENT NO. : US4348550 A

Inventors  : Frank C. Pirz, Lawrence R. Rabiner, Aaron E. Rosenberg, Jay G. Wilpon

Publication date: Sep 7, 1982

Filing date: Jun 9, 1980

A speech controlled dialing circuit identifies input utterances which may be a command word, repertory word (dialing name or number) or non-recognized ("Other") word.

Responsive to the identification of each occurring input utterance, a set of predetermined templates are selected to identify the next occuring utterance.

A programmed microprocessor system is described to implement the main controller function.

Automatic and repertory dialing arrangements permit telephone system subscribers to access frequently called telephone numbers

without time consuming and errorprone manual dialing.

Such dialers find widespread use in the business environment where efficient utilization of telephone communication is economically important.

While the number of manual operations needed to complete the dialing of a telephone number is significantly reduced 

in known repertory dialing systems, manual operations have not been eliminated.

A directory stores a set of dialing signals corresponding to the repertory words.Responsive to each input utterance, a speech analyzer produces a signal representative of the acoustic features of the utterance. 

Jointly responsive to the stored template signals and the utterance acoustic features signal, a spoken word recognizer generates a signal identifying the input utterance. 

Upon the identification of an input utterance as one of the repertory words, the corresponding dialing signal is retrieved from the directory store.

 Responsive to the identification of an input utterance, a set of predetermined template signals are selected to identify the next occurring input utterance.

According to one aspect of the invention, a control signal corresponding to each utterance identifying signal is generated. 

Signals to address templates in the template memory are produced responsive to the next occurring input utterance. 

Jointly responsive to the utterance identifying control signal and the memory addressing signals, a predetermined a set of template signals are applied to the 

spoken word recognizer.

Detection of other than the two short interval command words "error" or "stop" causes a dialing signal corresponding to the repertory name to be retrieved from a directory store.

 The dialing signal is then supplied to the user telephone line. Recognition of the short utterance as "error" resets the dialer to recognize the succeeding utterance as one of the repertory word templates.

 Detection of the shortened utterance as "stop" returns the dialer to its rest mode. After the dialing signal is obtained, the dialer is switched to its call state and the recognizer is conditioned to detect an input utterance as the isolated word template "hang up." 

FIR FILTER DESIGN

                       FREQUENCY SAMPLING METHOD

The frequency-sampling method for FIR filter design is a direct technique imaginable when a desired frequency response has been specified.

It consists of uniformly sampling the desired frequency response, and performing an inverse DFT to obtain the corresponding (finite) impulse response .

By using N-point filter response, the continuous frequency response is calculated as an interpolation of the sampled frequency response.

The approximation error would then be exactly zero at the sampling frequencies and would be finite in frequencies between them.

The smoother the frequency response being approximated, the smaller will be the error of 

One way to reduce the error is to increase the number of frequency samples.

FIR Filter Design

                           WINDOWING METHOD

FIR Filters have poles at origin. Thus, FIR filters are always stable. The order of FIR filters is given by 'N-1' where 'N' is the length of h(n).

FIR Filters can be designed by two ways 1) using a Window function 2) By Frequency Sampling method.

Focusing on the method of using a window function, a window function is a mathematical function that is zero-valued outside of some chosen interval.

Choice of the window function depends on the stop band attenuation required.

A scilab code for designing a High pass FIR filter with Hanning window was written and executed. 

Chebyshev Filter Design

                                 

In this experiment, Low pass and High pass Chebyshev filters were designed. By giving the values of pass band attenuation (Ap), stop band attenuation (As), pass band frequency (Fpass), stop band frequency (Fstop), sampling frequency (Fs) as inputs Low Pass and High pass filters were designed.
As a result order of the filter, cutoff frequency, normalised H(s), denormalised H^(s) and transfer function H(Z) was obtained.
The magnitude response and the pole zero plot was obatined. The filters we designed had ripples in the pass band and was monotonic in stop band indicating it was Chebyshev I filter. Also the ripples obtained verified the order of filter.

Butterworth Filter Design

                                 

In this experiment,  Butterworth filters were designed using Scilab. By giving the values of pass band attenuation (Ap), stop band attenuation (As), pass band frequency (Fpass), stop band frequency (Fstop), sampling frequency (Fs) as inputs Low Pass and High pass filters were designed.
As a result order of the filter, cutoff frequency, normalised H(s), denormalised H^(s) and transfer function H(Z) was obtained.
The pole-zero plot and magnitude response were observed. From the magnitude response it can be concluded that Butterworth filter is monotonic as it does'nt have any ripple in its pass band and stop band. 

Since Butterworth Filter has a flat response in its pass band, it finds application in audio processing.

OAM and OSM

Overlap Add method and Overlap Save method

FFT is known for its fast computational applications one of which includes fast convolution of signals than the usual direct method.Overlap Add and Overlap Save are the two block processing techniques used for fast convolution.Here the zeros are appended to the signal or filter sequence until the input signals are of equal length.

 In OAM,  the output of a long sequence can be calculated by simply summing the outputs of each block of the input. Here the output is such that overlapping takes place in the tail of the output from the previous block and the beginning of the output from the present block.

In OSM, instead of sectioning the input and then calculating the output, we decompose the input by adding some preceeding values to each of the input to make it equivalent to L = length of  x[n] and then do circular convolution with h[N] and finally eliminate the zero padded  values from each convoluted output.

Fast Fourier Transform

Fast Fourier Transform

Fast Fourier Transform (FFT), as the name suggests is faster than DFT since computations in FFT are done parallely. So here the signal is divided into 2 parts, so the computation is carried out only for N/2 signal values.

 C language code of DITFFT (Decimation in Time FFT)  algorithm was executed for the FFT calculation. In DITFFT, the input is divided as even and odd parts also it is not in order while the output is obtained in order.

FFT has made working in frequency domain computationally feasible in image processing and signal processing because of its application in large integer and polynomial multiplication.

Discrete Fourier Transform

Discrete Fourier Transform

The limitations of Discrete Time Fourier Transform (DTFT) are overcome by Discrete Fourier Transform (DFT). DFT evaluates enough frequency components of the signal that are required for its reconstruction and hence is the frequency sampled version of DTFT.

 In this experiment, we made use of arrays to store the real and imaginary parts of the signal.  Also the count of number of Real Addition and Real Multiplication for N=4 and N=8 was verified.

By manipulating the signal in different ways, it was observed that expansion of signal in time domain results in compression in the frequency domain.

The major contribution of DFT is in the area of spectral analysis of systems.

Convolution and Correlation

Convolution and Correlation

To perform and verify the results of the theoretical concepts in Digital Signal Processing, it is very important to have a good command over a programming language like C.

Through the first experiment, output of Convolution ( Linear, Circular and Linear Convolution using circular) and Correlation (Auto and Cross) codes were verified with the results of mathematical method.

Convolution is a method of combining two signals and gives an output signal. So two signals of different length were taken for Linear and Circular convolution and aliasing effect was observed in case of Circular convolution. Application of Convolution is to find the output of the system.

Correlation is a measure of degree of similarity between two signals. We found the Auto-Correlation and Cross Correlation of original, delayed and scaled signal.