Speech rhythm is one of the main prosodic features of spoken languages. The quantitative research data on this topic in the literature are not numerous may be due to the confusing divergence in this concept definition. In order to render clear this concept, we investigated a speech segment in Mandarin. Due to its delicate temporal structure, it leads easily to the semantic ambiguities, well known in speech recognition. In order to understand how human manage the speech meaning in a natural language processing, we designed a series of lexical duration and pause interval modulations based on the speech segment for production. By analyzing the resultant acoustic patterns, we observed two types of temporal grouping. The type 1 is context dependent, the time range has no regularity, the modulations suffered from semantic ambiguities. The type 2 includes the stable patterns presenting temporal regularities. We consider the temporal structures in this type as speech rhythm. This definition corresponds to the articulatory periodicity that produces the consonant and vowel alternations. They are context independent and reproducible.