How much does transcription cost per hour of audio?

It depends entirely on who is doing it. Human transcription is normally priced per audio minute and an hour of audio costs a meaningful amount of money. Automatic transcription is typically bundled into a monthly subscription where an hour costs a small fraction of that, and several built-in tools cost nothing.

How much does transcription software cost?

Most consumer transcription apps are sold as a monthly or annual subscription with a usage allowance, typically in the low single-digit to low double-digit dollars a month. Some charge per hour of audio processed instead. Free tiers are common but usually capped by length or number of files.

Is AI transcription cheaper than human transcription?

Dramatically, yes — the difference is more than an order of magnitude. The trade is accuracy on difficult audio. For clean speech the gap in quality is now small; for poor recordings, heavy accents or overlapping speakers, a human still wins clearly.

Is there any way to transcribe audio for free?

Several. Apple's Voice Memos transcribes recordings on recent iPhones, YouTube auto-captions uploads, and Zoom, Teams and Google Meet all produce transcripts for meetings you host. For a lot of ordinary needs these cost nothing and are sufficient.

Why does medical or legal transcription cost more?

Specialist vocabulary, formatting requirements and confidentiality obligations. A transcriptionist working on clinical or legal material needs domain training and often certification, and the work carries compliance overhead that general transcription does not.

How Much Does Transcription Cost? Human vs AI vs Free

Q: Why is human transcription so expensive?

Because it is priced by the transcriptionist's time, not yours. A skilled transcriptionist takes roughly four hours to produce a clean transcript of one hour of clear audio, and considerably longer for poor audio or several speakers. You are buying those hours.

The short answer: There are three pricing models, not one. Human transcription is priced per audio minute and reflects a person’s working hours. Automatic transcription is priced as a monthly subscription and costs a small fraction of that. Several built-in tools cost nothing at all — and for a lot of ordinary needs, that third option is the right answer.

Search for transcription pricing and you get a wall of per-minute rates with no explanation of why any of them are what they are. The rates make sense once you know what you are actually buying in each case.

A note on numbers: this page describes how transcription is priced and what drives it, rather than quoting current rates from named services. Pricing in this category has moved fast and repeatedly, and a specific figure written today would be misleading within months. Check the current price with the provider.

Model 1: human transcription, priced per audio minute

A human transcription service quotes per minute of audio, not per minute of work — and the distinction is where the cost comes from.

A skilled transcriptionist working on clear, single-speaker audio takes something in the region of four hours to produce a clean transcript of one hour of recording. That ratio worsens quickly with difficulty. Overlapping speakers, background noise, unfamiliar accents, technical vocabulary or poor microphone placement can push it to six or eight hours per audio hour. That is what you are paying for, and it is why the price per minute varies so much between quotes for what sounds like the same job.

Things that push a human quote up:

More speakers, particularly if they interrupt each other
Turnaround time — same-day work costs more than a week
Verbatim level — a strict verbatim transcript that captures every stammer and false start takes longer than a clean read, and the difference between full, clean and intelligent verbatim is worth settling before you ask for a quote
Specialist domain — medical and legal work needs domain-trained transcriptionists
Timestamps and speaker attribution, if requested at a fine granularity

Human transcription is not overpriced. It is a skilled person’s day, sold by the audio hour.

Model 2: automatic transcription, priced as a subscription

Automatic speech recognition has almost no marginal cost per minute once the model exists, so it is rarely sold by the minute to consumers. It is sold as a subscription with an allowance — a number of hours a month, or a cap on file length.

This is why AI transcription pricing looks strange next to human pricing: you are not buying labour, you are buying access. The economics are those of software, and a provider can afford to include far more audio than any individual will use.

Our own app follows this shape and we will state it plainly rather than being coy: three transcriptions are included free with no card, and Pro starts from $2.99 a month with memos up to 90 minutes. That is our current pricing, and it is on the pricing page rather than buried here.

Model 3: free, and more capable than most people realise

This is the section most articles on this topic skip, because most articles on this topic are selling something.

A large share of what people pay to transcribe can be transcribed for nothing:

Apple’s Voice Memos transcribes recordings on recent iPhones, at no cost, on the device you already own.
YouTube auto-captions almost every upload and shows the transcript in a panel — see transcribing a YouTube video to text.
Zoom, Teams and Google Meet all produce a transcript of meetings you host, included in the plan you are already paying for. Transcribing a Zoom, Teams or Meet recording starts there for exactly this reason.
Apple Podcasts and Spotify carry transcripts for a large part of their catalogues.

If your need is “I want to search this recording for the bit where they said the number”, free is very likely enough, and you should not spend anything.

So what should you actually pay?

The honest decision rule is about consequences, not features.

Pay nothing when the transcript is for you, the audio is clean, and an error costs you a re-listen. Notes to self, lecture recordings, searching a meeting you attended.

Pay for software when you transcribe regularly, need speaker attribution, or want the transcript to arrive without you doing anything. The recurring cost is small and the time saved is real. This is also the only tier where speaker labels are routinely included, and for interviews and multi-person meetings that is the difference between a usable document and a wall of text.

Pay a human when the transcript is evidence. Legal proceedings, regulatory submissions, published research where the exact words are the data, or audio so poor that automation genuinely fails. Also when accuracy failure is expensive — a misheard dosage or clause is not a re-listen.

Most people asking this question are in the first two rows and are being shown pricing for the third.

Cost is not the only axis

Two things routinely matter more than the price and get less attention.

Time. How long transcription takes varies from near-real-time for automatic services to days for human work with a standard turnaround. If you need it this afternoon, the cheapest option that meets the deadline is the one to buy.

Accuracy on your audio. Published accuracy figures are measured on clean studio speech. How accurate AI transcription actually is explains why the number falls apart in a real meeting room — and a cheap transcript you have to correct extensively is not cheap.

The summary

Model	Priced by	Best for
Human service	Audio minute	Evidence, poor audio, specialist domains
Software subscription	Month, with an allowance	Regular use, speaker labels, hands-off
Built-in tools	Nothing	Personal notes, hosted meetings, clean audio

The gap between the first and third rows is enormous. Work out which row you are in before you compare prices within one.