Basic Operations On DSPP

 We used code composer studio to program the board. The board we used in this experiment was DSP kit  TMS320F28335. For implementation of algorithms we used C language. Basic operations were performed on the board such as arithmetic, logical, bit manipulation, etc.  DSPP's are generally used for a specific application rather than general purpose processors. They are fast processors due to Harvard Architecture, pipelining and parallel processing techniques. 

Patent Review: Adaptive noise reduction filter with continously variable sliding bandwidth

Patent no 6,154,547

The patent provides method for dynamic noise reduction of an audio system. It compares average audio signal entering with that of a leaving signal. Adaptation rate is decided by requirement of filter bandwidth. The patent shows drawbacks of another patent (patent no 3,889,108) which reduces the noise using predefined known or estimated power value. The drawback of this is thatthe phase difference gives larger error signal. The low pass filter used has adjustable coefficients which produces variable cut off frequency. A comparator and multiplexer are used to provide required signal which then used in LMS adaptation block where adaptation parameter gets decided. This adaptation delta generates adaptive filter coefficients. Thus if audio signal is of low frequency range then the cut off frequency reduces and hence for high frequencies. Delay and attack signal are used for reducing and increasing the bandwidth. It also sets higher cut off frequency which is the limit of adaptation delta. The patent claims dynamic reduction of noise, adapting steps, adaptive filter coefficients, attack and decay time constant, upper and lower limit of cut off frequencies. Added to these it also claims about monitoring of quality, averaging of input and output signals, error calculation using multiplied average value. Thus the overall patent gives experimental design for dynamic noise reduction in audio reproduction and gives the advantage of each aspect of the design which improves the quality of audio at minimum requirement of energy.

IEEE Paper Review: Obtaining speech assets for judgement analysis on low-pass filtered emotional speech.

In this paper, different types of available assets for emotional research are reviewed. It gives brief idea on methods used to obtain the current corpus composed of high quality spontaneous speech. The paper further proposes an experiment that uses low pass filtering, on natural speech data, in order to investigate the impact cue masking has on listener’s perception of emotion in speech. The paper explains Brunswick Model For Speech. Acoustic properties are measured in reference to the encoding process and perception test on emotional speech are considered as decoding aspects. The paper further reviews existing emotional speech data. In this section, main type of data which are simulated, natural and induced vocal expressions are discussed. The fourth part gives idea about how the assets are obtained through few experiments. Two participants, placed in two isolation booths (soundproof booths), that were asked to perform a cooperative based task. Meanwhile researcher monitored, manipulated, and recorded the procedure. Later in cue masking experiment, effects of filtering on tonal quality of speech are studied. It shows that tone of the conversation can often be clearly heard even though filtering of certain frequencies are there. Thus the listeners are still able to infer vocal affect from natural speech. Then the author gives rating strategy for emotional speech. It suggest two affective scales, namely evaluation and activity on five point Liker-scale. The conclusion of this paper is that it gives research on Brunswick lens model, methods for obtaining high quality natural speech through mood induction procedure. It proposes an experiment to isolate semantic content from acoustic information by masking cues using low pass filters. The filtered and original signal are rated and compared on scale of activity and evaluation.

EXP 8: FIR Filter Design Using FSM

In the previous experiment FIR filter was designed using window function which is little bit complicated method. Another method used for FIR filter design was using Frequency Sampling Method(FSM). While performing the experiment, this method was found to be simpler than the method using window function. In this desired frequency response was specified. The given frequency response was sampled at a set of equally spaced frequencies to obtain N samples.Thus , it gave the N-point DFT of Hd(2pnk/N). Thus by using the IDFT formula, the filter co-efficients can be calculated. Increase in number of samples reduces the error.Thus we FIR filter design using Frequency Sampling Method and noted the output.

 

EXP 7: FIR FILTER DESIGN USING WINDOWING METHOD

The FIR filter was designed using Windowing method. The method multiplied window function to desired impulse response to obtain h(n). Window function had to be selected by an user and then other input parameters such as attenuations and frequencies had to be given. Out of five known window function, Hanning window function was used in the experiment. The length of the impulse response was equivalent to the order of filter.  After finding h(n), Z- transform was used to find transfer function H(z).

EXP 6: Chebyshev Filter Design

The next design was of Chebyshev filter. Again Scilab was used. The code was very similar to that of Butterworth filter, just few changes had to be done such as formula for calculating analog filter parameters. Input parameter such as attenuation and frequency for pass band band stop band had to be given with sampling frequency. Bilinear Transformation was used for converting Laplace to z domain. No ripples were observed in stop band which is the specialty of this filter further no of ripples in pass band gave the oder of the filter.

EXP 5: Digital Butterworth Design:

The design of digital butterworth filter was done on an open source tool - Scilab. Its functions are same as that of MATLAB. The transfer function of the filter was calculated in the Laplace (s) domain and then converted to the z domain by using the Bilinear Transform Method. Both low pass and high pass filter were designed. Order for each of the designs was higher than tens. The theoretical calculation for filter design seems to be easier than the code but the program can design a filter of any order.