What Speaker Identification Costs and What Moves the Price

When people ask what speaker identification costs, I start by explaining that it is not a flat add-on. A single interviewer and a single guest can be labelled almost as you transcribe. A panel of six with a moderator is a different job, because every line needs a decision about who is talking before the words are even checked.

That decision takes listening time. For each turn, the person preparing the captions has to match a voice to a name, confirm it against a program or a roster, and keep the label consistent from the first minute to the last. In the course we teach this as its own skill, and the same is true of pricing: it reflects effort, not just minutes of footage.

So a quote for labelled captions usually sits above a quote for plain transcription of the same running time. How far above depends on a handful of factors, which the rest of this note walks through.

The first driver is headcount. Two speakers who take long turns are cheap to label. Five speakers who trade short remarks every few seconds are not, because each change needs a label, a line break and a check that the right name sits on the right words.

Switch rate matters even more than the raw number of people. A lecture with one presenter and a short question session at the end is mostly a single voice. A roundtable where people finish each other's sentences can mean several label changes a minute, and that adds up across an hour of recording.

When you ask for a quote, a rough count of speakers and a note on whether the conversation is relaxed or fast will tell the preparer more than the running time alone. It is one of the first things I look at when I estimate the work.

• Few speakers with long turns: lowest effort
• Many speakers with short turns: highest effort
• Moderator plus audience questions: somewhere between

Overlap is the single feature that most often surprises people. When two people speak at once, the captions have to choose between showing both lines, marking the interruption, or favoring the main speaker. Each choice takes judgment and a replay of the passage, sometimes several.

Audience questions add their own cost. A person asking from the floor, far from a microphone, may be faint, and you may not know their name. We would label them in a consistent way, such as an audience member, and that needs a style decision agreed before the work starts.

Recordings with a lot of crosstalk or off-mic questions take longer per minute, and the price follows. If your event has a question session, tell us whether the questions are audible and whether names are available. That detail can change the estimate noticeably.

Poor audio makes every part of captioning slower, but it hits speaker identification hard. Voices that sound alike through a bad recording are hard to separate, and a thin or echoing room makes it worse. The preparer ends up replaying the same few seconds to be sure who is speaking.

Names are the other half. If you send a speaker list with spellings, job titles and the order of appearance, labelling becomes a matter of matching voices to a short list. If you send nothing, the preparer must work out names from introductions in the recording, and that takes time.

A clean recording with a supplied roster is the cheapest case. A noisy room with unnamed speakers is the dearest. Most real recordings fall somewhere between, and a short sample of the audio lets us place yours more accurately than a guess.

• Clear audio with a roster: lowest cost
• Clear audio, names unknown: moderate
• Noisy audio, names unknown: highest

Style decisions shape the work too. Do you want full names, first names or roles such as moderator and panelist? Should the label appear on every line or only when the speaker changes? Should a continuing speaker be marked again after a long pause? None of these is difficult, but each should be settled up front.

Changing the style halfway through is where costs creep in. Relabelling an hour of captions because the client now wants surnames instead of first names is real work, and it is better avoided by agreeing the format at the start.

Deadlines add a final layer. A tight turnaround means the work cannot wait for a quiet day and may require prioritising it over other jobs. We tell you what a rush does to the estimate before you commit, so there are no surprises when the quote arrives.

The simplest way to get an accurate figure is to describe the recording honestly. Give the running time, the number of speakers, how often they overlap, whether names are known, how clean the audio is and when you need the files. That is the same set of facts I use when I plan a job.

If you are unsure, send a short extract, a few minutes taken from the busiest part of the event. It tells us more than a long description, and it lets us flag problems such as a faint microphone before they become delays.

What you should not expect is a single per-minute price that holds for every video. Two recordings of the same length can differ widely once speakers, overlap and audio are taken into account, and a fair quote reflects that difference.

• Running time and speaker count
• Overlap and audience questions
• Audio quality and available names
• Preferred label style and deadline