At published list prices, batch speech-to-text from the major cloud APIs typically runs about $0.10 to $0.60 per hour of audio at the entry tier with no volume commitment, with most providers between roughly $0.15 and $0.36. That is around a cent a minute or less. Real-time streaming of the same audio usually costs more, sometimes about five times more, and add-ons plus the engineering around the API often cost more than the transcription itself.
This page normalizes prices to one unit, dollars per hour of audio, shows the arithmetic for three common workloads, lists the costs that are not in the rate, and sets out what you agree to when the audio goes to a provider. If the recordings are your own (meetings, interviews, lectures, dictation), the last sections show how to skip the meter entirely by transcribing on your iPhone or Mac.
How speech-to-text APIs bill
The unit is duration of audio, not file size and not words. Everything else is a modifier.
- Per second or per minute, rounded up. Most providers round each request up to the next second; some apply a minimum per job, such as 15 seconds, so a two-second voice note is billed as fifteen. If your workload is thousands of short clips, that floor matters more than the headline rate.
- Batch and streaming are different products. Pre-recorded, asynchronous transcription is the cheap tier. Streaming costs more, and some providers bill streaming by how long the connection stays open rather than how much audio you send, so an idle open microphone still meters.
- Channels can multiply. Some providers bill each audio channel separately, so a two-channel call recording costs twice its length.
- Add-ons stack. Speaker labels, word timestamps, language detection, custom vocabulary, redaction of personal data, summaries and translation are often priced on top. Speaker labels tend to add a few cents per hour; translation and summaries can add tens of cents per hour.
- Commitments and volume tiers. Most providers publish lower rates for committed volume, sometimes a fifth or more below pay-as-you-go.
- Free credits run out. Introductory credits and small monthly free allowances are for evaluation, not a plan.
Typical price ranges
English, entry paid tier, no commitment, converted to dollars per hour of audio. Prices checked on 30 August 2026; cloud pricing changes often, so treat these as ranges, not quotes.
| What you buy | Typical list price per hour of audio |
|---|---|
| Batch transcription, entry model | About $0.10 to $0.60, most between $0.15 and $0.36 |
| Batch, larger or premium model | About $0.20 to $1.00 |
| Real-time streaming | From the same as batch to about five times more, up to about $1.00 |
| Speaker labels add-on | A few cents |
| Translation or summary add-ons | Roughly $0.10 to $0.70 |
Most providers sell more than one model, and the larger one is what you will probably end up using, so treat the entry figure as the floor.
Three worked examples
One person dictating
30 minutes a day, 5 days a week, 52 weeks: 30 x 5 x 52 = 7,800 minutes, or 130 hours a year.
- At $0.15 per hour of streaming: 130 x 0.15 = $19.50 a year.
- At $1.00 per hour of streaming: 130 x 1.00 = $130 a year.
The trap is session billing. If your app holds the connection open two hours a day to catch half an hour of speech, you are billed for 2 x 5 x 52 = 520 hours, and at $0.15 that is $78 rather than $19.50. The raw API cost is small; you pay the difference in building and running the dictation tool around it.
A team recording meetings
10 one-hour meetings a week, 52 weeks: 520 hours a year.
- At $0.15 per hour: $78 a year.
- At $0.60 per hour: $312 a year.
- Add speaker labels, which meeting transcripts need, and the low end rises to around $88.
The API returns a transcript, not a service. Recording the meeting, storing the audio, search, sharing, permissions and support are all yours to build.
A 500-hour back catalog, once
- At $0.10 per hour: $50.
- At $0.36 per hour: $180.
- At $0.60 per hour: $300.
A one-off archive is where cloud pricing is at its best: it finishes overnight because the provider runs it in parallel, and you never touch the infrastructure again.
The costs that are not in the per-hour price
- Storage and egress. If you keep the audio in the cloud, you pay storage every month and, on some platforms, a fee to download it again. Recordings dwarf transcripts in size.
- Building the integration. Authentication, uploads, callbacks or polling, format conversion, splitting long files, and mapping the response into your own data. Days, not hours.
- Errors and retries. Rate limits, timeouts on long files, dropped streaming connections, and jobs that fail after you were billed for the attempt.
- Monitoring. Someone has to notice when the queue backs up, when accuracy shifts after a model update, and when a model is retired.
- Keys and access. API keys leak. Rotating them and keeping audio storage private is ongoing work.
- Ownership. All of it needs a name attached. A pipeline nobody owns is an outage waiting for a quiet week.
What you are agreeing to when the audio goes to an API
This is a summary of patterns across providers' published terms, not legal advice. Terms differ and change, so read the current version, and involve your own advisor for anything regulated. If you work under professional confidentiality obligations, medical and legal dictation covers the questions those reviews raise.
- Training and service improvement. Some providers may use customer audio to improve their services unless you opt out; at least one has offered a discount in exchange for letting it log your data.
- Retention. Abuse-monitoring logs are commonly kept for a period such as 30 days, and zero-retention options often require approval or an enterprise contract.
- Region. Processing may happen in a region other than the one you call, unless you pin a region where that is offered.
- Contract over website. For some vendors, the retention and training terms that apply to you are the ones in your agreement, not on the public page. Ask for them in writing.
The practical questions are the same everywhere: where is it processed, how long is it kept, is it used for training by default, can that be turned off, who are the subprocessors, and can you pin a region. On-device versus cloud transcription covers the wider trade-off.
When a cloud API is the right answer
Sometimes it is: a large archive with a deadline, server-side pipelines over recordings that already live in your own storage, a language your devices do not handle, or hardware too old to run a speech model. Self-hosting sits in between and has its own bill, covered in should you self-host a transcription server.
Or skip the meter: transcribe on your own device
For recordings you make yourself, the per-hour question has a simpler answer. Recent iPhones and Apple silicon Macs can run a speech model on the device, so there is nothing to upload, no connection to bill and no integration to maintain. The cost is the device you already own and the app.
ThinkScribe works this way. It sets up the speech model on your device for you, transcribes after you stop recording, and works offline. There is no per-minute bill: it is free to download with free uses of Pro features, and one subscription covers iPhone, iPad, Mac and Apple Watch (there is also a one-time purchase). There is no account, and your recordings and transcripts are never sent to us.
How to transcribe without a per-minute bill in ThinkScribe
- Install ThinkScribe on your iPhone, iPad or Mac.
- Choose the spoken language: the language chip on the iPhone Record screen, or Settings > Language for the default. There are 28 languages.
- Record a meeting, lecture or note, or bring in a file: Import Audio File on iPhone, Import File... (⌘I) or drag and drop on a Mac, where several files at once is fine.
- Read the transcript with speakers separated automatically; rename a speaker once and the voice is recognized in new recordings.
- Summarize it for a summary with action items, or ask questions about the recording.
- Export as text, subtitles, JSON, Markdown, PDF, Word or CSV.
The complete voice to text guide and the meeting transcription use case show where to go from there.
Frequently asked questions
How much does speech-to-text cost per hour of audio?
At published list prices checked in August 2026, batch transcription from the major cloud APIs typically costs about $0.10 to $0.60 per hour of audio at the entry tier with no commitment, with most providers between $0.15 and $0.36. Larger models cost more, up to about $1.00 an hour, and add-ons such as speaker labels or redaction are charged on top.
Why is streaming transcription more expensive than batch?
Batch work can be queued and scheduled when the provider has spare capacity, so it is cheap to serve. Streaming holds a connection and a slice of compute open for the whole session and must return partial results within a fraction of a second. Some providers also bill streaming by connection time, so silence is billable.
What are the hidden costs of a transcription API?
Storage of the audio, building and maintaining the integration, retries after failed jobs, monitoring for model changes, key management and someone to own it. For a small team these usually cost more than the per-hour rate itself.
When does self-hosting a speech model beat paying per minute?
On hardware you already run, the extra cost is mostly electricity, so heavy use is cheap. A rented GPU left running costs several hundred dollars a month whether it is busy or idle, so it only wins at high volume. You also take on the security and upkeep the API price was covering.
Can I transcribe recordings without paying per minute?
Yes. If the recordings are on your iPhone, iPad or Mac, ThinkScribe transcribes them on the device, with no per-minute bill and no account. It is free to download with free uses of Pro features, and your recordings and transcripts are never sent to us.
