AssemblyAI
by AssemblyAI
Getting-started guide
Turn recordings into speaker-labelled text — a transcription service built for programs to call, with a browser playground for trying it on your own audio first.
Current prices: Universal-2 $0.15/hr · See all plans
Last reviewed 2026-09-08 · for AssemblyAI
You sign in, drag an audio or video file into the browser playground, and a transcript comes back with each speaker's lines separated. You flick switches to change what comes back, run the file again, and compare. When it looks right, you copy the code the page has written for you into your own project.
AssemblyAI is a speech-to-text service that programs talk to through an API (a way for programs to talk to each other without a person clicking): your code sends a recording, and it sends back the words, with timestamps, speaker labels and — if you ask — a summary. There is no consumer app and no meeting bot. It is the transcription engine other products are built on, sold by the hour of audio.
So it is a component, not a destination. Speaker separation, translation, topic tagging and blanking out personal details are switches on one request rather than separate products, and the model (the AI "brain" that does the actual thinking) it routes to is chosen the same way.
You do not need to write code for the first hour. Signing up gets you a browser playground where you upload a recording, flip features on and off, and read the transcript on screen. Anything past that — a batch of files, an app of your own — means code. If you only want your own meetings transcribed, a meeting-notes tool does that with no code at all.
New accounts get a free credit grant with no card asked for — a one-off grant rather than a monthly allowance, which AssemblyAI's FAQ says does not expire. Everything after that is billed by the hour of audio rather than by subscription, with live transcription as someone speaks costing roughly double the flagship model's rate. Current rates are on its Launchpad page. Playground runs bill the same way.
Free accounts cannot opt out of model training. The setting that stops AssemblyAI using your files to improve its models is, by its own docs, a paid-plan feature. Evaluate with audio you would not mind a stranger reading.
AssemblyAI's security page says audio and transcripts are not stored after processing. Its own documentation is more specific and less absolute: with no special agreement in place, uploaded audio starts being deleted after about 24 hours and transcripts after around 30 days. Neither is instant, and the docs are the version to plan around.
Uploads a file in the browser and shows the transcript, speaker by speaker.
Why this one: Playground runs draw down the same credit as production and appear in the dashboard's cost tab, so what you spend evaluating is a number you can multiply — not a free demo priced differently from the real thing.
Tweak: Use a difficult recording: crosstalk, an accent, background noise. A clean file only confirms what you assumed.
Uploads the same recording again in the playground with the speaker-label and language-detection switches turned on, and compares the two transcripts.
Why this one: Every capability here is a flag on one request rather than a second service to buy, so re-running one file with switches on and off is how you size the job — and the playground then writes the code for the exact combination you chose.
Runs the same audio through Universal-2 and Universal-3.5 Pro and compares both.
Why this one: Universal-2 costs noticeably less per hour than Universal-3.5 Pro — invisible on one file, decisive at volume, and a quality gap only your own audio can settle. The free credit covers both runs; rebuilding on the other one later will not be free.
Blanks names, phone numbers and card numbers from the transcript, and can bleep them from the audio in the same request.
Why this one: On a free account this is the only privacy control you actually have, because the training opt-out is paid-only. Redacting as you transcribe keeps identifying details out of a file you cannot opt out of.
Want the full picture first? See the AssemblyAI overview for pricing, the free tier and related tools.
