AssemblyAI Documentation

Quickstart	Description
Pre-recorded Speech-to-Text API	Transcribes recorded audio and video asynchronously across 99 languages, with support on speaker diarization, PII redaction, and more features. Built for teams processing recorded calls, podcasts, or meetings at scale.
Streaming Speech-to-Text API	Handles live audio over WebSocket with ~150ms p50 latency, powering voice agents, live captions, and real-time agent-assist tools in contact centers.
Text-to-Speech API	Turns text into speech and streams the audio back as it is synthesized, over one WebSocket session per conversation or a single HTTP request. Thousands of voices plus voice cloning, in six languages, built for voice agents that need the first audio of each reply fast.
Synchronous Speech-to-Text API	Transcribes audio up to 20 minutes long over HTTP (up to 120 seconds through the SDKs or when uploading while recording) in a single request-response call — no polling or webhooks needed. It’s suited for quick, low-latency transcription use cases with need for limited transcription features.
Dictation API	Transcribes audio up to 20 minutes long over HTTP (up to 120 seconds through the SDK or when uploading while recording) and formats the text with an LLM in the same request. You include formatting instructions with your audio, and the transcript comes back already cleaned up and formatted.
Speech Understanding API	Layers speech analysis on top of transcripts — summarization, sentiment, topic detection, and more over audio — for extracting structured insight.
Guardrails API	Provides safety, privacy, and compliance controls — PII handling, content moderation, and related safeguards — for teams that need to govern what audio data flows through their pipeline.
LLM Gateway API	Gives unified access to frontier language models through the same API, so teams can apply reasoning models to their transcription without a separate integration.

Quickstart

Description

Transcribes recorded audio and video asynchronously across 99 languages, with support on speaker diarization, PII redaction, and more features. Built for teams processing recorded calls, podcasts, or meetings at scale.

Streaming Speech-to-Text API

Handles live audio over WebSocket with ~150ms p50 latency, powering voice agents, live captions, and real-time agent-assist tools in contact centers.

Text-to-Speech API

Turns text into speech and streams the audio back as it is synthesized, over one WebSocket session per conversation or a single HTTP request. Thousands of voices plus voice cloning, in six languages, built for voice agents that need the first audio of each reply fast.

Synchronous Speech-to-Text API

Transcribes audio up to 20 minutes long over HTTP (up to 120 seconds through the SDKs or when uploading while recording) in a single request-response call — no polling or webhooks needed. It’s suited for quick, low-latency transcription use cases with need for limited transcription features.

Dictation API

Transcribes audio up to 20 minutes long over HTTP (up to 120 seconds through the SDK or when uploading while recording) and formats the text with an LLM in the same request. You include formatting instructions with your audio, and the transcript comes back already cleaned up and formatted.

Speech Understanding API

Layers speech analysis on top of transcripts — summarization, sentiment, topic detection, and more over audio — for extracting structured insight.

Guardrails API

Provides safety, privacy, and compliance controls — PII handling, content moderation, and related safeguards — for teams that need to govern what audio data flows through their pipeline.

LLM Gateway API

Gives unified access to frontier language models through the same API, so teams can apply reasoning models to their transcription without a separate integration.

Choose how you build

API reference

Build directly against the raw HTTP and WebSocket API for full control over requests and responses.

SDK

Use the Python or TypeScript SDK to integrate transcription, streaming, and voice agents directly into your own codebase.

Build with an AI coding agent

Give Claude Code, Cursor, or any MCP-compatible coding agent our live docs (docs MCP server) and skill, so it writes correct code.

AssemblyAI MCP server

Connect Claude, Cursor, or any MCP client to your account, then ask it to transcribe, summarize, and analyze audio for you.

Integrations

Drop AssemblyAI into existing voice-agent frameworks and platforms — LiveKit, Pipecat, Twilio, Langflow, and others.

Getting started

Build & Evaluate

Administration

Trust & security

AssemblyAI Documentation

Choose how you build

API reference

SDK

Build with an AI coding agent

AssemblyAI MCP server

Integrations

​Choose how you build

API reference

SDK

Build with an AI coding agent

AssemblyAI MCP server

Integrations

Choose how you build