The raw data for the analysis are the waveforms as recorded directly with the help of the appropriate Sound Forge software. Several trials were conducted before selecting the reliable waveformsfor further analysis.
Musical notes are expected to be highly periodicas compared to noise. The analysis of noise requires stochastic methods whichfollow their own methodology. On the other hand, analysis of periodical signalsis a much simpler affair. Therefore, it was felt necessary that the currentwaveforms should be checked for periodicity. Short time windows of thewaveforms were selected and amplified. It was seen that in all cases theamplified signal displayed periodicity (Fig.3). Hence, it was felt that nostochastic analysis was called for.
The energy – frequency spectra indicate thedistribution of the energy of the wave form among the various frequencies ofthe periodic signal. The lowest frequency is the fundamental frequency and thehigher ones are the sub harmonics. These frequencies are also called “formants”. It is seen from all thespectra that the energy expressed in terms of decibels decreases withincreasing frequencies. Since the spectra (Fig. 4 - 10) use the logarithmicscale, it is easy to estimate the frequency at which the energy falls to 1% ofits value at the fundamental frequency. This helps us in identifying theeffective formants.

 first formantdepends on whether the voice is that of a male or a female. Male voices usuallyhave a lower pitch than female voices and hence a lower fundamental frequency.Usually, the ratio of the two pitches is 1:2.
The information contained in the fig. 4 -10 is presented in a complementary form in the figs. 11 and 12. These show thefrequency – time spectra with the energy level as a parameter. These are calledspectrograms and form in a sense the“signature” of the sound. They do notadd anything to the conclusions but present a three dimensional display of theinformation and contain in themselves the entire information which can beextracted from the wave forms.
The number of formants depends upon theresonance produced in the voice box. Male singers are capable of producinghigher resonance than female singers. This is a well recognized fact which hasalso been demonstrated by earlier worker1. This is indicated by thelarger number of formants for the male voices as compared to the female ones.For instance, in the present case the number of formants for all the malevoices were 5(sa), 3(ri), 6(ga), 5(ma), 5(pa), 6(dha) and 2(ni). Thecorresponding values for the female voices were 3, 2, 3,3,3,5 and 1. It shouldbe noted in this context that Indian music is predominantly nasal especiallyfor female voices, unlike western music which is sung with the throat. This isthe reason for the lower number of formants for the female voices as comparedto the male voices. This, of course, requires a deeper study.

The information provided above helps incharacterizing the sound pattern. It is a simple matter for an experienced person to be able to identify from this the musical note being produced and whetherthe voice is that of a male or female. However, if a specific voice needs to beidentified more detailed investigations are needed which form part of thescience of Speech Recognition.