Working With User Research Sessions Recorded in Other Languages

Where translation fits in a multilingual study

International studies are usually run in one of three ways: a native-speaking moderator runs sessions in the participant's language, a researcher moderates with a live interpreter, or participants who speak a shared language take part in that language. Only the first two produce recordings the core team cannot follow directly, and they produce different material.

With a native-speaking moderator, the recording is entirely in one language and needs a full translation for the team. With an interpreter, the recording contains both languages, often overlapping, and the interpreter's version is already a summary filtered through one person. Translating the participant's original words afterward lets you check what the interpreter conveyed, which matters when a finding rests on nuance.

Either way, translation turns hours of recordings into text the whole team can read, search and code. The task is to do that without losing the participant's meaning or exposing their identity.

Consent that covers transcription and translation

Participants consent to specific uses of their recording. Make sure your consent form and introduction cover:

• That the session will be recorded, transcribed and translated, and that software tools may be used for this.
• Who will see recordings, transcripts and translations, and whether quotes may appear in internal reports.
• How long recordings and transcripts will be kept, and how participants can withdraw.
• Whether recordings may be processed by third-party services, and in which countries, if your organization requires that disclosure.

Data protection and privacy laws, such as the GDPR in the European Union, can apply to research recordings, and requirements differ by country. Your research operations, legal or privacy team should approve the consent wording and tools. This is general information, not legal advice. If the original consent did not mention translation or external tools, ask before processing the recording.

Anonymization before anything is shared

Session recordings contain personal data: names, employers, locations, account details read aloud, faces and screens. A good default is to keep raw recordings in a restricted location and share only cleaned transcripts.

• Assign participant codes such as P1 and P2, and replace names in transcripts, including names of family members and colleagues mentioned in passing.
• Remove or generalize identifying details: specific employers, small towns, unusual job titles.
• Check translated transcripts separately. A name the speech recognizer spelled oddly can slip past a search for the correct spelling.
• If the session was a screen recording, remember the picture may show email addresses or account data even when the audio does not.

Where possible, export only the audio for transcription and translation, which keeps faces and screens out of the process entirely.

Analyzing translated transcripts

Translated transcripts make synthesis possible for the whole team, but they need some preparation and a few habits.

Machine transcripts do not say who is speaking. Before coding, mark the moderator's questions and the participant's answers, which is quick in a one-to-one session with a clear question and answer pattern. The article on speaker diarization explains why this step is still manual in many tools. For group sessions, the separate article on focus group transcription covers multi-speaker material.

Keep the original-language transcript alongside the translation, with timestamps, so any line can be traced back to the recording. Code themes on the translation, but treat hedging, politeness and intensity with care: translation tends to flatten them. A participant who says, in effect, "it is perhaps a little difficult" may be expressing strong frustration in a culture where direct criticism of a product is unusual. A native-speaking researcher or moderator should review how such statements are coded.

Be cautious about counting. Words and phrases do not map one to one between languages, so frequency counts across translated transcripts in different languages are rarely comparable.

Verifying key quotes before they reach a report

Quotes carry weight in research reports, and a mistranslated quote can push a product decision in the wrong direction. Before a translated quote goes into a report or presentation:

1. Find the timestamp and listen to the original audio around it, including the question that prompted it.
2. Ask a fluent colleague, the moderator or a professional translator to confirm the translation and the tone.
3. Check that removing surrounding context does not change the meaning.
4. Mark it clearly as translated, for example "P4, translated from Japanese," so readers know it is not verbatim.
5. Remove any identifying details from the quote itself.

Transcripts for analysis are often cleaned of filler words, which is usually fine; the trade-offs are covered in verbatim vs clean verbatim transcription.

A workflow from recording to report

1. Confirm that consent covers transcription, translation and the tools you plan to use.
2. Store raw recordings in a restricted location and export the audio track for processing.
3. Produce a transcript in the original language and a translated transcript for each session.
4. Mark moderator and participant turns, and replace personal details with codes.
5. Share only the cleaned transcripts with the wider team for coding and synthesis.
6. Have a native-speaking moderator or researcher review coding of nuanced or culturally loaded statements.
7. Verify every quote that will appear in a report against the original audio.
8. Delete recordings and transcripts according to your retention policy and consent terms.

Pitfalls and limits of machine-translated research data

• Think-aloud sessions often contain quiet, fragmented speech that is harder to recognize; see transcribing quiet audio for recording tips.
• Overlapping speech between moderator and participant, or with an interpreter, reduces transcript accuracy in the overlapping parts.
• Product names and interface labels may be misheard or translated when they should stay as they appear in your product.
• Sessions where participants mix languages, such as English product terms inside Japanese sentences, can confuse recognition and translation.
• Machine translation cannot tell you what a participant meant beyond their words. Cultural interpretation remains the job of researchers who know the context.

How mydubly fits a research workflow

mydubly takes a recorded audio or video file uploaded in the browser and returns a transcript in the spoken language, with a timestamped version, and optionally a translated transcript, in 21 languages. It also produces subtitles as SRT or VTT, which some teams find handy for watching a session clip with translated subtitles. The spoken language is detected automatically. See the audio translator for audio files and audio to text for transcription alone.

It does not label speakers, so marking moderator and participant turns stays manual. It does not anonymize transcripts, and it processes recorded files only, not live sessions. Files can be up to two hours long. For privacy, only compressed audio chunks are sent over HTTPS, a video file itself stays on your device, results are deleted within 30 minutes of completion, and the privacy policy says user data is not used for training. Whether that fits your consent terms and data policy is for your team to decide. As a cost example, a 60-minute session costs 60 credits (6¢) for the transcript and translated transcript, so ten such sessions cost 600 credits (60¢).

Your next study

Before your next international round, update the consent form to cover transcription and translation, agree who will verify quotes in each language, and decide where raw recordings will live. Then process one pilot session end to end, from audio export to verified quote, to check the workflow before the full study. The research interview transcription page and the audio translator describe the transcription and translation outputs.