A common setup in community-based Latino health research: bilingual interviews with first, second, and 1.5-generation participants, conducted in English or Spanish depending on each participant's preference, at community organizations across a Midwestern city. The recordings come back mixed. Some are fully in Spanish, some fully in English, some switching mid-interview as participants reach for the language that fits the concept they're describing. One transcriptionist flags the mixed recordings as too difficult. Another delivers transcripts where heritage speaker constructions have been silently corrected to standard Spanish. The researcher is now reviewing every transcript manually before analysis can begin. This is the predictable result of sending US Hispanic research recordings, whether from one-on-one interviews, focus groups, or Zoom, Teams, and Webex sessions, to transcription services built for standard, monolingual speech.

Focus groups generate some of the most demanding audio in qualitative research. Six to twelve people talking, sometimes over each other, sometimes in a room with bad acoustics, sometimes over a Zoom call with background noise from a home environment. Getting that audio into a clean, usable transcript is where a lot of research budgets and timelines get tested. Not every transcription service handles this well, and the right one often depends on the kind of focus group you're actually running.

A research team returns from fieldwork across São Paulo, Recife, and Porto Alegre. Three cities, three clearly distinct accents, two weeks of interviews. Back at the institution, someone books a Portuguese transcriptionist. Nobody specifies which variety of Portuguese they need. The transcripts come back with Nordestino expressions normalized to São Paulo usage, a participant whose name appears in three different spellings, and no timestamps. Technically, the words are mostly right. As research data, the transcripts are close to unusable. This happens because transcription for field research in Brazil requires decisions that general transcription services don't prompt researchers to make.