Interview Transcription Services for Research, Media, HR and Legal Work
Interview transcription converts a recorded interview into written text that can be quoted, coded, disclosed or filed. Espresso Translations transcribes research interviews, journalistic interviews, HR and grievance meetings, police, legal and insurance interviews, market research and podcast interviews from $5 per minute of audio, in 150+ languages, and a second person always checks the finished draft with the recording running. Speaker labels, timestamps, verbatim or clean read and anonymization are part of the work rather than extras. Work takes 2 to 3 working days as a rule, 24 hours where it is urgent, and a free quote reaches you inside the hour.
Human Transcribers, Not Speech Recognition
From $5 Per Minute of Audio
Verbatim or Clean Read
Speaker Labels and Timestamps
150+ Languages
Free Quote in 1 Hour


Send Your Interview Recording for a Free Quote
A project manager replies within 1 hour
What are interview transcription services? Interview transcription services turn a recorded interview into a written transcript that can be read, quoted, coded, disclosed or filed. The recording can be a research interview, a journalist’s interview, an HR or disciplinary meeting, a police or insurance interview, or a podcast; what makes the transcript usable is the same in every case. Each speaker is labeled and the label never changes, timestamps tie lines back to the audio, overlapping speech and inaudible passages are marked rather than guessed at, and the style — strict verbatim or clean read — is decided before the typing starts. Espresso Translations transcribes interviews from $5 per minute of audio in 150+ languages, with a second person checking every transcript against the recording, standard delivery in 2 to 3 working days, 24-hour express for urgent work, a signed confidentiality agreement on every file, and a free quote within 1 hour.
What Do Interview Transcription Services Actually Deliver?
An interview transcript is the only part of the interview anyone else will ever read, so what the service delivers is not a text file but an accurate record somebody can rely on. Recording and transcription are two different problems, and the fastest way to transcribe your interviews is usually not to transcribe them yourself. Two voices have to be told apart through overlap and interruption. Times have to be attached so a reader can go back to the audio. A passage nobody can make out has to be marked as inaudible rather than quietly completed. And the decision about whether hesitations, repetitions and false starts stay in or come out has to be made before the work starts, because it is not a formatting option afterwards.
The thing that separates a usable interview transcript from a raw dump is that it is checked against the recording rather than read for sense. A transcript that reads smoothly and silently repairs an interviewee’s grammar has changed what they said, and in a disciplinary hearing, a court file or a research dataset that is not a stylistic matter. Here a transcriber types the file and a second person plays the audio back against it, which is where the errors that read perfectly well get caught. Accurate transcripts come from that second pass; accurate text that nobody checked against the audio is a guess with good punctuation.
What arrives: the transcript in DOCX, RTF or plain text, speaker labels applied identically from the first line to the last, timestamps where you asked for them, overlapping speech marked as crosstalk, inaudible passages marked with their position in the recording, a note at the top of anything we could not resolve, and the names and terminology list built while working. The file is editable, shareable and searchable, so you can edit the transcript, circulate it or quote from it without going back to the audio at all. Accurate transcriptions come with that note attached; the note is the most useful line in the file and the one an automated tool will never write.
Who buys it: universities and research teams conducting research who have interview data to code, for whom the interview is the data collection step, journalists and producers working to a deadline, HR teams and employment lawyers who need a defensible record of a meeting, police forces and insurers with recorded interviews to disclose, market research agencies running depth interviews, and podcasters publishing the text alongside the episode. The work is the same work. What differs is who is allowed to hear the file and what the transcript has to withstand afterwards.
Which Interviews Do Clients Send Us?
Five kinds of interview cover almost everything that arrives, and the differences between them are real rather than marketing categories. The table is how a quote gets scoped: what the interview is, what its transcript has to carry, and what comes back.
| Interview | What the transcript has to carry | Usual deliverable |
|---|---|---|
| Research and academic interviews | Verbatim speech where the method asks for it, participant codes rather than names, inaudible passages timestamped, consistent labels the analysis software can autocode | DOCX or RTF laid out for qualitative analysis software |
| Journalistic and documentary interviews | Quotable accuracy, a timestamp at every answer so a line can be found in the audio, names and places checked against your spellings | Timestamped DOCX, or a Q and A document for publication |
| HR, grievance and disciplinary interviews | Everything said, by whom, in order, with nothing smoothed over; the chair, the employee and any companion separated; date and attendees recorded | DOCX formatted as a meeting record, timestamped throughout |
| Police, legal and insurance interviews | Strict verbatim with hesitations and repairs left in, every speaker identified, times carried through so the transcript can be checked against the original recording | Verbatim DOCX or PDF with a continuous timestamp convention |
| Market research and podcast interviews | A clean read that can be published or quoted, moderator and respondent separated, product and brand names spelled your way | Clean-read DOCX, plus SRT or VTT where the episode is published with captions |




What Has to Be Decided Before an Interview Is Transcribed?
Five decisions, and all five are cheaper to make now than to discover later. None of them is an upsell; they are the questions a project manager will ask you in the first reply.
Verbatim or Clean Read
Verbatim transcription records speech as it happened: repetitions, false starts, the ums and the you-knows, audible pauses, laughter. Clean read drops the filler and the stumbles and keeps the meaning. Neither is more accurate than the other; they answer different questions. Research into how something was said, anything that may be disputed, and any interview that could be disclosed need verbatim. Market research write-ups, published interviews, podcast show notes and most HR summaries read better clean, although an HR record of a contested meeting should not be cleaned at all.
Decide before the work starts, because this is not a format you can switch afterwards. Filler can be stripped out of a verbatim transcript later; it cannot be put back into a clean one without transcribing the recording again, and that is charged again. If you are unsure, ask for verbatim. If you are producing the transcript yourself rather than buying one, the layout conventions and a worked example are in our guide to the transcript of an interview.
Speaker Labels When Two or More People Talk
Speaker labels are what make an interview transcript usable rather than merely accurate. Send the convention you want — Interviewer and Interviewee, initials, real names, role labels such as Chair and Employee, or your own participant codes — and it is applied identically throughout. Consistency matters more than the choice, because anything that reads the file afterwards matches on the label string: one stray spelling splits a speaker in two across a dataset, and in a transcript going to a tribunal it looks like carelessness.
Where speech overlaps we mark it rather than choosing a winner, and where a passage is genuinely unintelligible we mark it with its time so you can go back to the recording and decide. Where two voices genuinely cannot be separated — two similar voices on one phone microphone, a group interview recorded from one end of a table — we label them consistently and say so in a note at the top of the file, instead of attributing an answer to the wrong person.
Timestamps, and Where They Go
Timestamps are what let a reader get back to the audio, and where they go depends on what the transcript is for. At every speaker change suits a two-voice interview and is the most common request. At a fixed interval — every thirty seconds, every minute — suits a long monologue or an interview somebody will quote from. Both together suits legal and investigative work. The format is hh:mm:ss, and it is matched to the original recording rather than to any edited copy.
Say which before the work starts, because retrofitting timestamps onto a finished document means playing the whole recording again. There is no charge for timestamps, for the speaker convention, or for the confidentiality agreement; they are part of the transcript, not a tier.
Anonymization, Consent and Research Approval
Two words get used interchangeably and they are not the same thing. Pseudonymization replaces names with codes and you keep the key. Anonymization removes the identifiers and the route back to them. We produce either, applied as the transcript is typed rather than as a find-and-replace afterwards, and we flag the indirect identifiers a name swap misses: a job title inside a small team, a named street or clinic, an unusual diagnosis, an employer with one site.
Where the interview is research, a transcriber is a third party with access to identifiable data, and most institutional review boards expect the protocol to name that and to require a confidentiality agreement. On a funded project the contracts office usually wants the agreement on its own paper, which is fine. Where the recording carries protected health information, HIPAA treats a transcriber as a business associate and your institution will want the paperwork for it, so say so before sending the file. Where your participants are in the EU or the UK, the GDPR applies to the recording whatever state you are sitting in.
Confidentiality and Who Hears the Recording
Interview audio is usually the most sensitive material a client sends: a grievance naming a manager, an unpublished story, a witness account, a participant who agreed to talk on conditions. So there is no confidentiality tier to buy here, and nothing to negotiate. Uploads are encrypted rather than emailed, exactly two people have access — the assigned transcriber and the project manager — and both of them, along with the company, have signed before the recording moves anywhere. None of that depends on what you sent: a quality interview recording in clear audio and a handset file full of background noise are handled identically.
We will sign your own agreement or your data processing terms if your legal or contracts team needs its wording; what binds the transcriber is a signed contract rather than a line in a privacy policy. Source recordings and working files are deleted on request when the project closes. If a matter is contentious and you want the file handled by one named person throughout, ask — that is how it runs by default on HR and legal work.
What Does Interview Transcription Cost, and How Fast Is It?
Interviews are quoted on their runtime at a floor of $5 per minute of audio: a 48-minute interview is a 48-minute job whether the answers were long or the silences were. The things that lift a quote above the floor are all amounts of work rather than amounts of margin — the number of voices to keep apart, how much of the audio has to be replayed before it yields, strict verbatim with a notation sheet, anonymization as the typing happens, and the deadline.
Nothing else is billed. Labels are not an extra, the timestamp convention is not an extra, the confidentiality agreement is not an extra, and names are taken out as part of the typing rather than as a second service. Where a sample flattered the recording, you get a revised number before any work starts and not an unexplained invoice afterwards. A doctoral candidate spending a stipend and a funded program are quoted identically, because there is no academic markup here to discount. Affordable transcription at this end of the market comes from doing the work once.
Expect the transcript in 2 to 3 working days, or inside 24 hours on the express tier; an hour of transcription means sixty minutes of audio rather than an hour of somebody’s time, and the turnaround time is quoted on the whole file. Batch work — thirty depth interviews, a set of investigation meetings, a season of episodes — is scheduled before it starts, since one advertised figure cannot be true of a single 40-minute file and of forty hours of audio at the same time. Where a hearing, a copy date or a submission is the real constraint, name it in the first message and the answer will be straight.
More interviews cross a language boundary than any other kind of recording, which adds two things to many of these projects. The transcribing happens in whichever of 150+ languages the interview was conducted in (see multilingual transcription), and the finished transcript is then translated from $0.10 per word. That order is what ethics committees and editors both want, because the original transcript remains the record and the translation is merely the version people read. Where a document has to be certified for an official purpose, certified translation is $30 per page of up to 250 words.
Two points from the list above bear repeating. We do not bill for putting our own mistakes right. And the ISO 17100:2015 certification this company holds governs a translation process, not a transcription one — it is the reason a second linguist reviews every translation we deliver, and any agency presenting it as a transcription accreditation is misreading its own certificate.
| Work | Price | Turnaround |
|---|---|---|
| Transcription of a recorded interview | From $5 per minute of audio | 2 to 3 working days; 24-hour express for urgent work |
| Anonymized or pseudonymized interview transcript | From $5 per minute of audio | 2 to 3 working days |
| Translation of a finished interview transcript | From $0.10 per word | 2 to 3 working days |
| Quote | Free | Within 1 hour |
Additional Interview, Audio and Translation Services
An interview project rarely needs only one of these. Alongside interview work we provide transcription services for every other kind of recording, academic transcription for lectures, focus groups and fieldwork, human transcription where the point is that a person rather than a model produced the file, video transcription where the interview was filmed, legal transcription for depositions and case audio, and conference call transcription for recorded calls and meetings. Four language services sit around all of that: documents an authority will only take certified go through certified translation, captions and subtitles are subtitling work, a script that has to be spoken becomes voiceover, and anything belonging to a product entering another market belongs to localization.
How Does an Interview Transcription Order Run, Step by Step?
Four steps. Everything that determines how good the transcript is happens in the first one, before any money changes hands.
Step 1: The Recording, Plus Everything You Know About It
Upload the audio file or the filmed version in whatever your recorder made of it — MP3, WAV, M4A, AAC, WMA, MP4, MOV are all ordinary here — or send a link to the file on Google Drive, Dropbox or SharePoint, or to a Zoom, Teams or Meet recording. With it, send the briefing: who is in the room and in what order they first speak, what the interview was for, the participant codes or labels you use, the names and specialist terms already known to be in there, and whether anything has to come out of the transcript. A two-line email of that kind is worth more than a round of corrections later.
Step 2: Scope, the Four Questions, and a Quote Inside the Hour
Someone listens to a representative stretch, counts the audio minutes, and replies within the hour with a rate and a delivery date. The same reply puts four questions to you while they are still cheap to answer: verbatim or clean read, where the timestamps belong, what the speakers are called, and whether names are pseudonymized or removed outright. Agreements that have to be in place before the recording travels are exchanged at this point, on your paper if your contracts office requires it.
Step 3: Transcribing, Then a Second Pass With the Audio Running
The file goes to one of our transcriptionists who works in your kind of interview material, and when the draft exists a different person plays the recording against it from the beginning. Reading a draft for sense is not the same test and does not find the same faults: what the second pass catches is the technical term that was misheard into a real word, the answer credited to the wrong voice, the figure with two digits swapped, the surname spelled two ways in one file. Whatever remains unresolved reaches you as a numbered query.
Step 4: Delivery, the Query List, and Deletion
Your transcript arrives in the format you chose, headed by a short note: the time of every inaudible passage, every point of unresolvable overlap, and any voice we could not reliably separate. Spend five minutes checking the opening against the audio, because a convention applied the wrong way shows itself there and nowhere later. Any error of ours is put right as part of the work. Say the word when the matter is closed and the source recordings and working files are deleted.
Should an Interview Go to a Transcriber or to AI Transcription?
AI transcription has a real place in interview work and it is not the one the pricing implies. An automated pass is fast, cheap and good enough on one clear speaker in a quiet room, and it is a sensible way to find the ten minutes of a two-hour recording you actually care about. An interview is almost never one clear speaker in a quiet room. It is two or more people, interrupting, on a phone, a video call or a handset on a meeting room table, often with an accent or a second language in play and always with vocabulary the model has not been given. Speech to text and voice typing are the same technology wearing different labels, and neither is able to transcribe an interrupted conversation reliably. That is what human-powered transcription is for, and why it sits in the workflow at the point where the material starts to matter.
What makes this matter more in an interview than in a voice memo is the way the failure arrives. Nothing in an automated transcript says “I was unsure here”: the model writes a fluent sentence the interviewee never produced, and that sentence goes on to be quoted in an article, coded into a dataset or read aloud at a hearing. A person leaves a marked gap. So the sensible arrangement is not either-or — automate the first pass over your own material if it helps you find what matters, then put the part that will be relied on in front of a transcriber. The two things we will not do are sell a machine draft as a finished interview transcript and sell transcription software or a self-serve tool, because neither is our business. It is also worth being plain about the difference, since the two things are sold side by side: a transcription app with a free plan, a dashboard and a subscription price is software you operate, and a meeting assistant that joins the call is automation of the recording step. A transcription tool and a transcription service answer different problems. And no accuracy percentage appears anywhere on this page for either route, because the honest number is a property of your recording rather than of a supplier.
What Automatic Transcription Gets Wrong in a Two-Voice Interview
Four failures, and they repeat. Two voices at once become one voice, or one becomes three — and in a good interview the voices are at once constantly. An accent or a second-language construction is quietly regularised into textbook English, which edits the meaning without leaving a trace. An unfamiliar word — a drug, a product, a statute, a surname — is replaced by a familiar one that sounds like it and reads convincingly. And because the software has no state for “could not hear that”, you are never shown where to check. Three of those four are judgments about what the recording means, which is why a better model next year does not solve them. They are also decisions a transcriber makes before they begin transcribing, not errors found afterwards.
Which Related Espresso Services Sit Around Interview Transcription?
Interview work sits next to several of these, and one of them may be the page you actually wanted. Keeping a project with one supplier is mostly about consistency: the labels, the names and the terminology stay the same across thirty interviews, which is what makes them comparable.
General Transcription Services
Audio or video that is not an interview — a meeting, an event, a lecture, a call? Start at our transcription services hub, where every other kind of recording lives.
Academic and Research Transcription
Focus groups, lectures, fieldwork or a viva alongside the interviews? See academic transcription, which covers the notation systems, analysis-software formats and review-board paperwork research brings with it.
Video and Human Transcription
Was the interview filmed, or do you specifically need it on the record that a person produced the file? See video transcription for footage, with SRT and VTT captions, and human transcription for the comparison with automated services.
Legal and Court Transcription
Depositions, hearings, recorded client meetings or case audio? See legal transcription and court transcription.
Certified and Multilingual Translation
An interview recorded in another language, or a transcript an authority will only accept as a certified document? Multilingual transcription covers 150+ languages and certified translation services are $30 per page of up to 250 words.
Subtitling, Voiceover and Localization
Publishing the interview as video or audio for another market? Captions and subtitles are priced from $5 per video minute, a spoken version is voiceover work quoted per project, and the wider program sits at localization.
2,000+
Linguists
ISO 17100
Certified quality standard
1 hour
Free quote
150+
Languages Supported
What Else Do Clients Ask About Interview Transcription Services?
How much do interview transcription services cost?
The floor is $5 per minute of audio, charged on runtime: 48 minutes of recording is a 48-minute job however much or little was said inside it. Above the floor the variables are the number of voices, how much replaying the audio demands, strict verbatim with a notation sheet, anonymization, and how soon you need it. Translation of the finished transcript is a separate line from $0.10 per word. Most transcription services online quote per audio minute and ours is no different; send the recording or five representative minutes of it and the quote is free and back inside the hour.
How long does it take to transcribe an interview?
Two to three working days as standard; 24 hours on the express tier. Batches get an agreed schedule rather than a published number, because one figure cannot describe a single 40-minute interview and thirty of them. If you are doing it yourself, reckon on several times the runtime for clean two-voice audio and well beyond that once people start talking over each other, because nobody can transcribe audio at the speed it plays and the hours go on rewinding rather than on typing.
Do you transcribe verbatim or clean read?
Either, and the choice depends on what the transcript is for rather than on your budget. Verbatim keeps the repetitions, false starts, hesitations and marked pauses: right for research into how something was said, for anything that may be disputed, and for any interview that could be disclosed. Clean read drops the filler and keeps the meaning: right for market research write-ups, published interviews and podcast notes. Decide before the work starts, because filler can be stripped out of a verbatim transcript later and cannot be put back into a clean one.
Will you label the speakers?
Yes, at no extra cost, and in whichever scheme you hand us — Interviewer and Interviewee, initials, real names, roles such as Chair and Employee, or your own participant codes. Whatever it is, it stays character-for-character identical down the whole file, because every tool that later reads the transcript matches on that string and one variant turns one person into two. Two voices that genuinely cannot be told apart get a consistent pair of labels and a sentence at the top of the file saying so, which is better than a confident misattribution.
Can you add timestamps, and where do they go?
Yes, in hh:mm:ss, and you choose the convention: at every speaker change, at a fixed interval such as every minute, or both. Every speaker change suits a two-voice interview; a fixed interval suits a long answer somebody will quote from; both suit legal and investigative work where any line may need checking against the original recording. Say which before the work starts, because retrofitting timestamps onto a finished document means playing the whole recording again.
Can you anonymize the interview transcript?
Yes, and pseudonymization and anonymization are different jobs. Pseudonymization replaces names with codes and you keep the key; anonymization removes identifiers and the route back to them. We apply either as the transcript is typed rather than as a find-and-replace afterwards, and we flag the indirect identifiers a name swap leaves behind: a job title in a small team, a named clinic or street, an unusual diagnosis, an employer with one site. Send your participant code list with the recording if you have one.
Is my interview recording confidential, and will you sign our agreement?
Yes, and the arrangement is deliberately narrow. The recording is uploaded over an encrypted connection instead of being emailed, and two people can reach it: the transcriber assigned to the file and the project manager. Both have signed, as has the company, before the audio moves. Your own agreement or data processing terms are signed instead where your legal or contracts team needs its own wording. Recordings and working files are deleted on request once the matter closes, and on contentious HR or legal work one named person holds the file throughout.
Can you transcribe a Zoom or Teams interview?
Yes. Send the recording file or a link to it — a Zoom, Teams or Meet recording is ordinary interview audio as far as the work is concerned. Two practical notes. If the platform produced its own automatic transcript, send that too: it is not usable as a deliverable, but it sometimes resolves a spelling. And if the platform recorded a separate audio track per participant, send those rather than the mixed file, because separate tracks make overlapping speech far easier to untangle.
What audio and video formats can you work with?
Anything your recorder or your phone produced, and a link rather than a file where the recording is large: MP3, WAV, M4A, AAC, WMA, FLAC and AIFF on the audio side, MP4, MOV, AVI, MKV and WMV where the interview was filmed, and shared links to video or audio recordings sitting on a drive or held by a video-call platform. There is no separate rate to convert audio or video — audio recordings and filmed interview recordings are both quoted per minute of audio. One technical note worth acting on: the untouched recorder file beats a compressed copy, because compression discards the frequencies that distinguish two quiet voices, and that is exactly what an interview turns on.
Can you transcribe interviews in other languages, and translate them?
Yes, in 150+ languages. An interview conducted in Spanish can be transcribed in Spanish and the transcript then translated from $0.10 per word, which is the arrangement most research committees and most editors want: the original transcript stays as the record and the translation is what gets coded or published. Transcribing straight into English from foreign-language audio is possible, but it merges two decisions into one invisible step and leaves no original to check against, so we recommend against it for anything that will be published.
How do you handle two people talking over each other?
By marking it rather than choosing a winner. Overlap is normal in a good interview and it is where automated transcription collapses, so it is the work that most justifies paying a person. A transcriber rewinds the passage, separates the turns where that is possible, and marks it as crosstalk where it is not. Two things help at the recording stage more than anything we can do afterwards: a microphone between the speakers rather than in front of one, and separate tracks per participant where the platform offers them.
Can you transcribe a police, insurance or investigation interview?
Yes, and that material gets the strictest treatment: full verbatim with hesitations and repairs left in, every speaker identified, and a continuous time convention so any line can be checked against the original recording. Nothing is summarized, corrected or filled in, and inaudible passages are marked with their time. Access is limited to one named transcriber and one project manager under a signed confidentiality agreement. What we deliver is a record of a recording; it is not a certification of anything said in it.
Is AI transcription good enough for an interview?
For finding what you need in a long recording, often. For a transcript anybody will rely on, no — and the reason is not the error rate but the way the errors arrive. An automated pass does not tell you which passage it guessed at: it writes a fluent sentence instead, and that sentence looks exactly like the ones that are correct. It also merges speakers as soon as they overlap and normalizes accents into standard English, which is a silent edit to what was said. Automate the first pass on your own material if it helps; send the part that matters to a person.
Who does the transcribing?
Transcribers who work in the kind of interview you are sending, picked by subject rather than by who happens to be free, each with five years of experience behind them and each signed up to confidentiality before a file is released to them. The draft then goes to a second person who plays the recording against it. Nobody here is polishing a machine output with a spell-checker; they are listening to your audio for as long as it runs, which is why four voices in a bad room cost what they cost. That is the difference between hiring a service and subscribing to an application.

