In speech, the sounds involved in an utterance are not produced independently of one another, but rather reflect the result of a complex process of sound concatenation. It is important to study these coarticulation effects in representations of speech signals since, for example, their cues can be helpful in the development of robust speech recognition systems. Representational tools, such as the spectrogram, are useful for visualizing spectral characteristics along the time axis; most of these tools are based on secondorder statistics, and it is interesting to consider other methods which might be useful in studying the problem of coarticulation. In recent years, sparse signal representations using suitable dictionaries of functions seem to provide an attractive alternative. With this alternative in mind, the present paper applies spectral and basis pursuit techniques to spanish synthesized signals. The results on a reduced vocabulary show that some prosodic and coarticulatory cues can be obtained from the basis pursuit method compared to the spectral representation.