Research Transcription That Preserves Your Data

Qualitative research stands entirely on the quality of its transcripts, and the way transcription goes wrong in research is subtler than an error rate. If a transcript smooths over hesitation, overlapping speech, or a participant correcting themselves, the analysis is working from evidence that has quietly changed - and the finding may be an artefact of the transcription convention rather than the data.

So the first question is which convention you need. True verbatim retains filler words, repetitions, false starts, and non-verbal sounds, which conversation analysis and discourse analysis treat as data. A cleaned transcript removes them for readability, which suits thematic analysis where meaning is carried by content. We transcribe to the convention you specify rather than applying a house style and letting you discover the difference during coding.

Focus groups are the hardest material in research transcription. Six participants with crosstalk is exactly where automated speaker attribution degrades, and an attribution error in a focus group transcript propagates straight into your coding. Transcripts come back with consistent speaker labels so NVivo, Atlas.ti, and MAXQDA workflows can use them directly.

Research recordings usually carry confidentiality obligations under an IRB protocol. Files are encrypted, access is logged, and material is never used to train a model. Worth deciding before you send files: whether identifiers should be removed by the transcriptionist, flagged for you to remove, or left intact in a secure transcript. All three are defensible; discovering nobody chose is not.

De-identification is harder in speech than researchers expect, because spoken data resists find-and-replace. Participants name people, employers, clinics, schools and neighbourhoods spontaneously, in forms that match no list prepared in advance. A transcript can contain no names at all and still identify someone to anyone who knows the setting - which, in a study of a specific community or institution, includes the readers most likely to see it. If a transcriptionist is doing the de-identification, they need a specification of what counts, because "remove identifiers" is not one.

Retention deserves the same attention. A protocol committing to destroy recordings at study end has to account for every copy, and third-party transcription creates copies by design. Ask how long the vendor keeps source audio after delivery, how long they keep the transcript, and whether deletion on request is confirmed. Grant and institutional requirements sometimes pull the other way, requiring preservation for a period after publication - and where that conflicts with a protocol commitment, the conflict is better surfaced to your IRB before signing with a vendor than after.

One practical note on timing and cost. Transcription is usually on the critical path between fieldwork and analysis, and it is the step most often underestimated in a project timeline. A study generating forty hours of interviews is a substantial transcription order, and at a $25 minimum per order, batching rather than sending files one at a time as they are collected makes a material difference to the total.