For assessing the validity of a morpho-typological classification of urban sounds with respect to sound assessment, factor analysis was applied to a database compiled in the city of Lyon and lead to a four-classes classification based on the number of lanes and whether a street is one-way or not. Free categorisation was then carried out on sound excerpts recorded in different sites corresponding to different classes. The main results are: a corpus of 42 h of five-channel recordings of a choice of sites from the classification at different times of the day; the validation of the relevance of the morpho-typological classification for perception; and a confirmation of the semantic oppositions between sources and background noise, as well as holistic and event sequences. Further, annoyance is controlled by the absence or the presence of human or nature sound sources, pleading for research toward automatic identification of sources. Last, categorisation confirms earlier investigations based on variance analysis: in narrow and large streets, traffic increase differently governs the increase of both noise level and annoyance.