In this paper, we present our work on speech-smile/shaking vowels classification. An efficient classification system would be a first step towards the estimation (from speech signals only) of amusement levels beyond smile, as indeed shaking vowels represent a transition from smile to laughter superimposed to speech. A database containing examples of both classes has been collected from acted and spontaneous speech corpora. An experimental study using several acoustic feature sets is presented here, and novel features are also proposed. The best configuration achieves a 30.1% error rate, hence well above chance.