It it is relative to the "current range". That is determined by a number of things. First, individuals have a certain range as a consequence of their anatomy. Second, languages can specify (in a social sense, not a strictly grammatical sense) that the range exploited in speech should be relatively high or relatively low. Variants of such whole-language positioning of the pitch frame of reference are found for various social groupings, e.g. high pitch may be a sign of status, or may be expected of females, and so on.
Third, there is usually a certain range of variation for stylistic / pragmatic functions, so that lowering the reference pitch might convey doubt or anger and raising it might convey excitement.
At this point we have narrowed down the range of pitch by reference to many semi-social and physical factors: we can call that the "register" in which tones are realized. A metaphorical way to look at "register" is that it is a box inside other boxes, and various rules (social or grammatical) determine how tall the box is, and where it sits inside the next box out.
One last factor regarding pitch range is that language have a phonetic tendency to continuously lower the box defining register, sometimes over the last few syllables but sometimes throughout the utterance. This lowering, known as downdrift, may be purely non-contrastive so that if we assign pitch numbers to H and L tones, a sequence like HLHLHLHL comes out as 9-6-8-5-7-4-6-3. In that case, H and L can be locally defined in terms of a current register value, then H is L+3. Initially, L i.e. the bottom of the register might be set at 6, but at every transition from L to H the register value decreases by 1.
However, this process of register lowering might be phonologically unpredictable (not just triggered by changing from L to H), and then we call it downstep.
There are, potentially, significant problems of comprehensibility arising in a language with highly-contrastive tone (where you have e.g. 5 tone levels and many minimal pairs so that "ma" on some pitch is potentially 5-ways ambiguous. However, once a person utters two syllables, the ambiguity is decreased, and within not much time at all, a listener can have established what a speaker's personal pitch range is. People may use visual cues if they can see, or other auditory voice-quality cues to make reasonable guesses about physiologically-based factors governing an individual's pitch range. Or they may just be confused for a moment, until they get used to the individual. Since the listener is likely to be first exposed to fixed expressions like "Hello, I am pleased to meet you. Please come in", the chances of actual confusion are diminished.
