**4. Methodology**

In this section, we present the algorithm implemented to find the frequency and amplitude of the first formants during any vowel-like segment. In order to analyse any speech fragment, a time-frequency analysis is needed. Short-time Fourier transforms (STFT), constant-Q [19] and wavelet transforms are some of the most commonly employed solutions in several systems. In this paper, the main idea is based on a previous work [20], tested on a large number of utterances produced by several different speakers; McCandless's discovery was found to be extremely successful. This algorithm is combined with some other ideas already developed by authors [11] in the context of polyphonic piano recordings.

We should remark that this manuscript comes to complement the work initiated in [1], so the recordings accepted by our system consists of only one vowel each, unlike the one presented in [20]. The latter developed a completely automatic algorithm which was meant to yield the first three formants during all voiced sounds in continuous unrestricted speech. For this reason, the algorithm developed in this paper can be implemented more easily and productively.
