AssemblyAI provides speech AI models and APIs that transcribe audio and video, understand what is said, and generate speech. Developers call them through an API or run the models in their own infrastructure.
The product includes pre-recorded and realtime speech-to-text, a sync endpoint that returns finished text for a short clip in the same request, and a speech understanding API that produces summaries, sentiment and entities from transcripts. Guardrails detect and redact personal information and unsafe content in voice data. A text-to-speech API built for voice agents, a dictation API and an LLM gateway are also listed.
The site is aimed at teams building voice agents, meeting notetakers, call analytics tools, dictation software and medical products. A medical mode adds clinical vocabulary for drugs, dosages and procedures, and a veterinary mode is available. A legal mode is listed as coming soon.
Models can be used through managed cloud inference with unlimited concurrency, or self-hosted in your own VPC or on-premises, which keeps control of your data with you. Speech-to-text covers 99 languages, and the latest recorded and realtime models handle 32 languages with code-switching.
The pricing page says you can start free and then pay as you go, with no commitments required. Custom plans are available through sales for larger volumes. The site names Zoom as a customer and lists developer documentation, an API reference, cookbooks and a changelog.
