JavaScript speech-to-text API

The speech-to-text API for JavaScript developers

Build Speech-to-Text into your app with a couple of API calls—transcribe audio files from a Node backend or stream live audio over a WebSocket, and get accurate text back with timestamps and speaker labels. Start free, no credit card.

Start building free

support-call.wav Done · 0:38

Speaker A 0:03

Thanks for calling—can you confirm the order number on the invoice?

Speaker B 0:09

Sure, it's the shipment that went out to Portland last Tuesday.

Speaker A 0:16

Found it. I'm issuing the replacement now—you'll get a confirmation email.

Word timestamps Real-time WebSocket Speaker labels

Plain JavaScript, no lock-in

Call the REST API with fetch from any runtime—Node, Deno, Bun, or the browser. A typed npm package is there if you want it, never required.

Real-time from any stream

Stream mic, telephony, or in-app audio over a WebSocket and partial and final transcripts come back within a few hundred milliseconds.

Audio files to text

Upload a file or point the API at a URL, then poll for the finished transcript—formatted text with timestamps in structured JSON.

36%

improvement in close rate

See case study

“The new Universal-3.5 Pro speech model from AssemblyAI is best so far in terms of accuracy, latency, and language switching.”

80%

increase in customer satisfaction

“Assembly has saved us countless hours managing models, and provided exceptional accuracy.”

36%

improvement in close rate

See case study

“The new Universal-3.5 Pro speech model from AssemblyAI is best so far in terms of accuracy, latency, and language switching.”

80%

increase in customer satisfaction

“Assembly has saved us countless hours managing models, and provided exceptional accuracy.”

Built for applications

A JavaScript API for speech-to-text in your app

Transcribe uploads in a Node batch job, power live captions in a meeting tool, or feed a voice agent from a telephony stream—the same API handles files and real-time audio. Every transcript comes back as structured JSON with speaker labels, timestamps, and language detection, so it drops straight into your product instead of a dictation demo.

Start building free See the live demo in the Playground

No credit card required

Streaming word error rate

Word error rate is the share of words the model gets wrong against a human reference—the standard measure of accuracy for live, realtime Speech-to-Text.

Normalized WER on the voice-agent caller corpus, mean of three noise conditions.

*Lower is better*

Voice-agent word error rate, realtime

AssemblyAI Universal-3.6 Pro Realtime

5.19%

Soniox Realtime

6.83%

ElevenLabs Scribe v2 Realtime

7.78%

Speechmatics Enhanced Realtime

8.14%

Google Gemini 3.5 Transcribe Live

8.48%

AWS Transcribe Streaming

8.58%

Deepgram Nova-3

8.64%

Source: AssemblyAI realtime benchmarks technical report v2026.09 (assemblyai.com/benchmarks), data 18 Aug–24 Sep 2026.

Common questions

: Send your audio to the API with a couple of fetch calls—submit a file or URL, then poll until the transcript is ready. For live audio, open a WebSocket and transcripts arrive as your users speak. The docs have copy-paste JavaScript for both audio files and streaming.
: You don't need one—the API works from plain JavaScript with fetch and a WebSocket, in any runtime. If you'd rather have a typed client, the assemblyai package on npm wraps the same API. You can also try it without writing code in the Playground.
: For live audio, compare realtime accuracy: in our realtime benchmarks technical report, AssemblyAI's Universal-3.6 Pro Realtime has the lowest voice-agent word error rate (5.19%), ahead of Soniox, ElevenLabs, Speechmatics, Google, AWS, and Deepgram. See the full benchmarks, then test the API on your own audio—that's the comparison that matters.
: Yes. Capture mic audio the way you already do and stream it to a speech-to-text API over a WebSocket. You get consistent results in Chrome, Safari, and Firefox—the Web Speech API's biggest limitation—plus control over the model and where your users' audio is processed.
: Yes. Node's built-in fetch and WebSocket are all you need—transcribe files in a batch job, stream live audio server-side, or power a voice agent. TypeScript projects can add the typed npm package for autocomplete on the response.
: Yes. Call the API from your server routes, or stream directly from the browser using short-lived tokens so your API key never ships to the client—no framework-specific package needed.
: AssemblyAI's free tier includes up to 185 hours of pre-recorded transcription and up to 333 hours of streaming—no credit card required. After that, pay-as-you-go pricing applies with no upfront commitments.