Audio and Video Transcription

A transcript is usually not an end product but an intermediate one. The text is later translated, summarised, turned into subtitles or added to a legal file. How the transcription should be done therefore depends on what will be done with it afterwards. An interview destined for qualitative analysis and a recording to be submitted to a court are not transcribed the same way.

When is it needed?

Transcription is asked for in four main areas. The first is research: in qualitative work, in-depth interviews are put into writing before they can be analysed. The second is corporate records: meetings, training sessions and webinars are turned into text to be archived or summarised. The third is media: interviews and podcasts are transcribed to prepare them for publication or for subtitling. The fourth is legal and official recordings.

What these areas have in common is that the quality of the recording decides the outcome directly. A clear recording with one speaker and no background noise is worked through quickly; a recording with people talking over one another, voices picked up at a distance and noise throughout takes many times as long. Recording conditions are, for that reason, the single biggest factor in estimating how long the work will take.

How the process works

Transcription is shaped by the decisions taken before any typing begins; try to change them later and much of the work has to be done again. The first steps in the flow are therefore not transcription but planning: what will be written, and how, is settled at the start. The quality of the recording bears directly on that plan – a clear recording speeds everything up, while a noisy one makes every step heavier. The sequence below applies the same framework to a short recording of a meeting and to a long research interview; only the level of detail and the rigour of the checking change, according to what the recording is for.

1. Assessing the recording. The clarity of the audio, the number of speakers and the running time are checked; the timescale and the price follow from these conditions.
2. Choosing the method. Verbatim, intelligent verbatim or edited; and whether time codes and speaker labels are wanted.
3. Transcription. The recording is put into writing; passages that cannot be made out are not filled in by guesswork but marked and left as they are.
4. Checking. The text is verified by a second person listening to the recording, or by listening through again; proper names and figures are confirmed separately.
5. Formatting and delivery. The transcript is laid out in the file format requested, with speaker labels and time codes, and handed over.

How it is delivered

A transcript is usually supplied as a plain document with the speakers labelled and, where requested, time codes added. If it is to become subtitles, the time codes are placed more frequently and at regular intervals; if it is going into a legal file, who said each thing and at what minute is stated explicitly. Where a translation of the recording is wanted as well, transcription and translation can be run as a single workflow.

What to watch for with this service

The problem that arises most often is how to handle passages that cannot be made out. In honest transcription nothing is filled in by guesswork; the inaudible part is marked and, where useful, noted with its duration. The owner of the recording can then complete those marked passages from their own knowledge of the context. An invented sentence is far riskier than a gap left open.

The second point is confidentiality. Audio recordings often carry personal data and private information: medical consultations, internal meetings, personal accounts. How such recordings are stored, who has access to them and what becomes of them once the work is finished should all be set out in writing. For matters of an official nature, the provenance and authenticity of the recording are a separate question; needs of that kind are best discussed in advance via our contact page.

Transcription and translation together

Transcription is rarely wanted on its own; when the recording is in a foreign language, or the output is needed in another language, transcription and translation become two steps of the same job. There are two routes here: the recording can be transcribed in its own language first and translated afterwards, or it can be rendered directly into the target language. Which is right depends on how the output will be used; if the recording is needed as a document in both languages, for instance, the two-stage route is the only option.

In bilingual work, time codes take on an added value. When you need to return to a particular moment in the recording later, the time stamp in the text anchors both the original and the translation to the same point. That matters especially in research and legal use, where a statement only carries its meaning together with its place in the recording. Planning the transcription and the translation together is far more efficient than trying to align them afterwards.

What determines the price

The cost of transcription follows not from the length of the recording but from how long it takes to work through a minute of audio. A minute of a clear recording may be written up in a few minutes, while the same minute of a noisy recording with several speakers can take far longer. A price therefore cannot sensibly be given without listening to the audio file; running time, clarity and the number of speakers are weighed together.

You can send us your recording and tell us what kind of output you need through the quote form. We explain how our pricing works on the pricing page; if your recording is also to be translated, you will find the languages we work in on our language pages.

Situations that come up often

• Poor recording quality. Recordings made at a distance, in noise, or with people talking over one another; nothing adds more to the time and the cost.
• An unspecified method. Not saying at the outset whether verbatim or edited is wanted; this means the work has to be done again.
• Confusion between speakers. Telling apart people whose voices are similar; where possible, a list of speakers and a voice sample should be shared beforehand.
• Specialised terms and names. Institution names, abbreviations and technical terms; a list supplied in advance improves accuracy.

This page sets out the general framework; the details of an individual case may differ. Before you decide, it is worth checking the institution’s own published information.