Investigating a tone language involves careful transcription of tone on words and phrases. This is challenging when the phonological categories--the tones or melodies--have not been identified. Effects such as coarticulation, sandhi, and phrase-level prosody appear as obstacles to early elicitation and classification of tone. This article presents open source software that can assist with solving this problem. Users listen to words and phrases of interest, before grouping them into clusters having the same tonal properties. In this manner, it is possible to quickly annotate words of interest in extended recordings, and compare items that may be widely separated in the source audio to obtain consistent labelling. Users have reported that it is possible to train one's ear to pick up on the linguistically salient distinctions. The approach is illustrated with data from Eastern Chatino (Mexico) and Alekano (Papua New Guinea).