AssemblyAI, Inc. builds speech AI models and sells them through APIs. Its main product is speech-to-text, also called automatic speech recognition (ASR): software that turns recorded or live speech into written text.
A program sends an audio file to the pre-recorded Speech-to-Text API, or streams live audio to the Realtime Speech-to-Text API, and gets back a transcript. The Universal-2 model supports 99 languages. Universal-3.5 Pro covers 18 languages, handles a change of language inside one conversation, and does speaker diarization, which means labeling which speaker said each part. The Speech Understanding API extracts speaker IDs, sentiment, chapters and summaries. Guardrails removes personal data (PII) from audio and transcripts. An LLM Gateway sends requests to language models such as GPT, Claude and Gemini through one endpoint. AssemblyAI also offers Dictation and Text-to-Speech APIs and a self-hosted option.
AssemblyAI is a hosted API. Use starts free and is then billed per hour of audio used, with extra per-hour charges for add-on features.