File Transcribe API vs a Raw Speech API (2026)

A raw speech API sells inference. You send audio, you get JSON, and you build the rest: storage, a status screen, speaker names, an SRT download, and a place to fix a wrong name.

The File Transcribe API sells the job the website already does. Your software sends the file. The transcript lands in the same library, with the same editor and the same TXT, JSON, SRT, and VTT export. Minutes come off the Plus balance.

If you are choosing infrastructure for a voice product, start with best transcription APIs compared. If you already have a recording and a script that should stop opening the upload page, this page is the split.

Quick pick

Official docs for the other two: AssemblyAI and Deepgram. Confirm their current prices and streaming claims on those sites. The numbers move.

What the File Transcribe API actually is

Plus only. You create a key in settings. The app prepares an upload or sends a public https link, then polls until the job is completed or failed. There is no webhook in v1.

The finished item is a normal library file. Open it when the transcript is good enough to read and one name is wrong. That is the reason to call this API instead of a raw one: you do not have to build the editor to get the editor.

Limits that matter: about 5,000 minutes a month on Plus, 2 GB and 3 hours per file, 90-day retention, 200 requests per key each hour. A Free or Pro key is rejected. The first file, step by step, is how to transcribe a file with the API. What shipped is the API is live.

What a raw speech API still does better

AssemblyAI and Deepgram are for product teams. You own the screens. They meter the audio. Streaming, and the extra speech models around it, are their job.

Pick them when:

• The product is live captions, a voice agent, or anything that cannot wait for a finished file.
• You will store the audio in your own bucket and you do not want a copy in someone else's library.
• Volume is large enough that a monthly seat with a minute cap is the wrong shape, and someone on the team will keep the integration working.
• You need that vendor's extra models, and you have read the current docs rather than a blog table.

File Transcribe will not pretend to be that stack. v1 does not stream. It does not send a webhook. It does not manage team membership. Ask AI stays in the product, not on the API.

The older three-way page is AssemblyAI vs Deepgram vs File Transcribe. It now includes the Plus key. It is still the page for "which buyer am I?"

When the Plus key is the whole integration

Use the File Transcribe API when the recording already exists and the next step is text or captions, not a new product.

A course library is a sheet of masters. An engineer can request an SRT per lesson and still open a bad one in the editor. The human checklist is how to caption an entire course library.

A sampled support call can leave the dialer as a WAV and come back as text. You are not embedding a speech engine in the helpdesk. You are skipping the upload click. The by-hand version is customer support transcription.

A server that already downloaded the MP4 should not open a browser. Send the file, poll, and save the SRT next to the master.

If nobody is going to write that script, do not buy Plus for the API. Upload on the homepage.

Price shape, not a rate-card fight

Plus is $49 a month. That includes the key and about 5,000 minutes. A raw API often looks cheaper per minute until you add the editor, the retries, and the person who keeps the integration up.

Do not treat a per-minute figure in an old post as a quote. Open the vendor's pricing page.

At a handful of hours a week, the seat is usually the smaller project. At very large volume, or at live audio, the raw API is the product you need, and File Transcribe is the wrong row. Pricing is the Plus side of that choice.

FAQ

Is the File Transcribe API a wrapper I should skip and call AssemblyAI myself?

Only if you want to build the library, the editor, and the caption download. If those are what you need, calling Plus is the shorter path. If you are shipping speech inside your own app, call a raw API and build your own screens.

Can I use both?

Yes. Plus for internal files and captions. A raw API inside the product you sell, when that product needs streaming or its own storage.

Does File Transcribe stream?

No. Send a finished file. Poll. If the audio is still being spoken, look at Deepgram or another streaming vendor.

Do I need a separate API plan?

No. The key is included with Plus. Pro stays the website: editor, captions, summary, translation.

Where are the exact calls?

API documentation. Keys live in dashboard settings.

Create the key on the Plus account that should be billed. Send one short file. Open it in the editor. Then decide whether the rest of the folder should go through the same door.

More guides

• Detect topics and keywords with AI
• AI sentiment and intent in transcriptions
• How AI transcriptions save time
• Test transcription accuracy
• Transcription guides

Further reading

• Google Speech-to-Text docs
• Zoom cloud recording help