Transcription is the process of producing a valid written record of an interview: would that it were so simple. It is easy to understand that the stage of data analysis is a selective, interpretive business. It is less easy, until you get down to the practicalities of the task, to appreciate that 'simple' transcription is itself a process of interpretation. Interestingly, the considerable improvements in voice recognition software since the mid-1990s have brought home to us just how interpretive 'straightforward' listening to the human voice is. As with narrative competence (see p. 48), listening in the active sense is a distinctively human characteristic whose precise operation is elusive, at least at the level of software reconstruction. Transcription, in a word, is a form of translation.
The most obvious loss is in the semantic properties of the human voice: those dimensions of speech (emphasis, pace, tone) which can radically alter what the words mean. That is easily appreciated in common-sense terms; reflecting them at the level of transcription is another matter. When the qualification of meaning is very clear, then some observation on the way the words were said (as in a play script) will be necessary. Incidentally, this interpretive qualification is one reason why the original tapes should be available as part of the chain of evidence.
Of course there are other features of speech which have a more specialized significance – hesitation phenomena, for example – another area of research for linguists. Analysis in terms of the minutiae of social interaction (as in ethnomethodology) is another dimension but less relevant to the analysis of the substantive detail of interviews.