AssemblyAI

Overview

AssemblyAI turns audio and video into text over a REST API and a streaming WebSocket, with official Python and Node.js SDKs. The pieces you combine:

• Pre-recorded transcription: submit a URL or file, the SDK polls until the job is completed or error. Models are chosen with speech_models (default ["universal-3-5-pro", "universal-2"], an ordered fallback list).
• Speech Understanding and Guardrails: options on the same request — speaker labels, sentiment, entities, key phrases, topics, content moderation, PII redaction.
• LLM Gateway: an OpenAI-style chat completions endpoint for summaries, chapters and questions about a transcript. It replaces LeMUR, which was shut down on 2026-03-31.
• Real-time STT: a WebSocket session (/v3/ws) that returns Turn events while audio is being spoken.

This skill targets Python SDK 1.x (1.6.1 at the time of writing). Version 1.0 removed every aai.Lemur* class and the microphone helper in assemblyai.extras; the old aai.RealtimeTranscriber (streaming v2) is gone as well.

Instructions

Step 1: Install and authenticate