Transcription and Translation of Audio and Video

Organisations record more than ever: interviews, meetings, webinars, customer calls, training videos, podcasts, focus groups and investigations. Recordings are hard to search, quote, review or translate until they become text, and many of them contain speech in more than one language.

This guide is for anyone commissioning transcription or translation of audio and video in the UK, whether a law firm, a researcher, a public body or a marketing team. It explains the outputs you can order, how the process works, what drives accuracy, and how to prepare files. Each specialist topic is summarised here and covered in depth in its own article.

What is transcription, and how is it different from translation?

Transcription is the conversion of speech in an audio or video recording into written text in the same language. Translation is the conversion of written text from one language into another. Turning a foreign-language recording into English text therefore involves both, either as two stages or combined into one.

A third service, interpreting, converts speech into another language in real time, for example in a meeting or interview. Interpreting leaves no written record unless the session is recorded and then transcribed. Our article on types of interpreting explains the spoken-language options.

Transcripts have value beyond the original recording. They let people search, quote, annotate and translate content, and they make audio and video accessible. The Government Digital Service's accessibility team puts it plainly: "Not providing captions or transcripts for videos excludes users who are deaf or hard of hearing."

What transcription and translation outputs can you order?

Transcription and translation outputs range from a plain same-language transcript to timed, translated subtitle files. Choosing the right output at the start avoids paying twice for work in the wrong format.

Output	What it is	Typical uses
Same-language transcript	Text of the recording in its original language, with speaker labels and optional timecodes	Meetings, interviews, podcasts, evidence, research
Translated transcript	A transcript translated into another language, either from a source transcript or directly from the audio	Foreign-language interviews, international meetings, media monitoring
Bilingual transcript	Original and translation side by side, usually timecoded	Legal review, qualitative analysis, quality checks
Captions	Timed same-language text with speaker identification and sound information	Accessibility of published video
Translated subtitles	Timed on-screen translation of dialogue and key on-screen text	Video for international audiences, e-learning, games
Summary or minutes	A condensed account of what was discussed, not what was said	Internal records where exact wording is not needed

A summary is not a transcript. If there is any chance the exact words will matter later, order a transcript and summarise from it.

How does the transcription and translation process work, and how should you prepare?

The transcription and translation process moves from file assessment to transcription, review, translation and final checks. Most delays and errors can be prevented at the first step, when the recordings and brief are prepared.

Assess the recordings. Confirm the total duration, number of speakers, languages and varieties, and audio quality.
Agree the brief. Choose the output, the transcription style, speaker labels, timestamps and file format.
Transcribe. A linguist who is a native speaker of the spoken language produces the transcript.
Review. A second pass checks difficult passages, names and terminology against the audio.
Translate, if needed. A translator works from the reviewed transcript into the target language, keeping segments and timecodes aligned.
Format and deliver. The text is delivered as a document, bilingual table, or timed caption or subtitle file.

Before you send recordings, work through this preparation checklist:

Send the original, highest-quality files rather than compressed copies or screen recordings.
List the speakers with names or roles, and say who speaks first if voices are similar.
Provide a glossary of names, places, products, acronyms and technical terms.
State the language and regional variety of each recording, and flag any mixed-language passages.
Say what the transcript is for: evidence, research, publication, captions or internal use.
Mark sections that do not need transcribing, such as long silences or off-topic breaks, with timecodes.
Explain any confidentiality, anonymisation or data protection requirements before files are shared.

Which file formats and timestamps should you ask for?

Ask for the format that matches the next step. Word or plain-text documents suit reading, editing and qualitative analysis software; a two- or three-column table suits bilingual review; and SRT or WebVTT files are the usual formats for captions and subtitles on video platforms.

Timestamps make a transcript verifiable. Common options are a timestamp at every speaker change, at fixed intervals such as every minute, or only next to inaudible passages. Agree the option in advance, because adding timestamps afterwards means listening to the whole recording again.

Which transcription style do you need: verbatim or clean verbatim?

The transcription style determines whether hesitations, false starts and non-verbal sounds are kept or removed. Full verbatim keeps everything as spoken; clean verbatim removes the noise of speech but keeps every meaningful word; an edited transcript rewrites speech into polished prose.

Evidence and detailed linguistic analysis need verbatim, while meetings, media and most research interviews suit clean verbatim. Our article on verbatim vs clean verbatim transcription compares the styles feature by feature, shows how Hansard's long-standing definition of a "substantially" verbatim report maps onto clean verbatim, and gives a checklist for briefing a transcriber on style.

How are recordings in other languages transcribed and translated?

Recordings in other languages are normally transcribed in the original language by a native speaker and then translated, although direct translated transcripts and bilingual layouts are also possible. The right route depends on whether the original wording may need to be checked later.

The need is common in the UK. The 2021 Census found that 91.1% of usual residents in England and Wales had English (or Welsh in Wales) as a main language, with Polish (1.1%, 612,000 people) and Romanian (0.8%, 472,000) the most common other main languages. Our guide to foreign-language transcription and translation compares the three output options in a table, explains how dialects and code-switching are handled, and covers UK GDPR restricted transfers when linguists work outside the UK.

What do legal, evidential and research recordings require?

Legal, evidential and research recordings need more than accurate typing: they need a style, layout and handling process that will stand up to later scrutiny by a court, an ethics committee or an archive.

Legal evidence and police interviews

In England and Wales, police interviews with suspects are recorded under PACE Codes E and F, and the recording itself is the primary record. The current PACE Code E applies to interviews carried out after 31 July 2018. Our article on transcription for legal evidence and police interviews explains why written records for prosecution files are often summaries, how to transcribe interpreted interviews in both languages, how section 139 of the Criminal Justice Act 2003 lets witnesses refresh their memory from a transcript, and how to order official court transcripts on form EX107. It also includes an eight-point checklist for evidential transcripts.

Court hearings are a separate case. According to GOV.UK, "Hearings at the Crown Court and at civil and family courts are always recorded", while magistrates’ court hearings are never recorded, and official transcripts are ordered through the court or tribunal rather than from a private provider.

Research interviews and focus groups

Research interviews must be transcribed consistently in a style that suits the analytic method, and participants must be protected. The UK Data Service advises that "Anonymisation should not be treated as a single technical step carried out at the end of a project." Our article on transcribing and translating research interviews matches transcription styles to methods such as thematic and conversation analysis, explains UK GDPR duties including special category data and processor contracts, and shows how to pseudonymise transcripts and translate interviews conducted in other languages.

Do videos need captions, subtitles or a transcript?

Videos with speech need captions for accessibility, translated subtitles for audiences who do not understand the language, and ideally a transcript as well. Audio-only content such as a podcast needs a transcript.

Under WCAG 2.2, Success Criterion 1.2.2 requires captions for prerecorded audio in synchronised media at Level A. UK public sector websites and apps must meet WCAG 2.2 AA under the accessibility regulations, although GOV.UK lists "pre-recorded audio and video published before 23 September 2020" among the things public bodies might not need to fix.

Private companies are not covered by those regulations, but the same standard is the usual benchmark for accessible marketing, training and product video. A single accurate transcript can serve several purposes at once: it becomes the source for captions, the basis for translated subtitles, and searchable text on the page. Our article on subtitles vs transcripts vs captions compares the three formats in a table, sets out the public sector rules and exemptions, explains why automatic captions need editing, and describes how translated subtitles are produced from a transcript.

What affects transcription accuracy, and is automatic transcription enough?

Transcription accuracy depends mostly on the recording itself, then on the number of speakers, the language and subject matter, and the skill of the transcriber. Automatic speech recognition can produce a fast first draft, but its output needs checking against the audio before it is relied on.

The main factors are:

Audio quality. Background noise, distant microphones, phone lines and heavy compression all cost words.
Speakers. Overlapping speech, similar voices and large groups are harder to attribute correctly.
Accent and language variety. Regional accents, dialects and mixed-language speech need a linguist who knows the variety.
Subject matter. Legal, medical and technical terms, names and acronyms are common error points without a glossary.
Style and extras. Full verbatim, frequent timestamps and detailed tags take longer to produce than clean verbatim.

Speech recognition tools are improving, but they do not flag their own mistakes, they misattribute speakers, and they handle accents, overlapping speech and less widely spoken languages unevenly. For material that will be published, analysed or relied on as evidence, use automatic output only as a draft that a human checks. Our article on machine translation post-editing discusses the same principle for translation.

Common mistakes when ordering transcription and translation

The most common mistakes when ordering transcription and translation are decisions not made before work starts. These cause the most rework:

Not stating the purpose. A transcript for minutes and a transcript for evidence are different products.
Ordering a direct translation of sensitive audio. Without a source-language transcript, no one can show what was actually said.
Using a bilingual colleague as transcriber. Fluent speakers tend to summarise and smooth over, which defeats the purpose of a transcript.
Translating unedited automatic captions. Every error in the source is carried into every target language.
Ignoring data protection. Recordings of identifiable people are personal data; check where files go and who can access them (see data security when outsourcing translation).
Changing conventions mid-project. Inconsistent speaker labels, tags or styles make a set of transcripts hard to compare or search.

How TranslationWorks can help with transcription and translation

TranslationWorks is a UK-based translation agency that transcribes audio and video recordings, from interviews and meetings to hearings, and translates the transcripts into 150+ languages, working remotely with clients everywhere. All work is carried out by native-speaking linguists, and TranslationWorks also provides interpreting, transcreation and custom language services. TranslationWorks is ISO 9001 (quality management) and ISO 27001 (information security) registered and Cyber Essentials certified, which supports the secure handling of confidential recordings, and has served 300+ companies. Read more about our audio and video transcription services and translation services, or request a free quote with your file details; quotes are sent within 24 hours.

Transcription and Translation of Audio and Video: A Guide

Key takeaways