AssemblyAI

AssemblyAI is a Voice AI infrastructure provider that supplies speech models and APIs for applications built around spoken conversations. AssemblyAI serves startups and enterprises working on recorded transcription, live voice interactions, and transcript analysis. The product combines speech recognition, speech generation, and speech understanding with managed or self-hosted inference.

AssemblyAI presents separate models for recorded speech, real-time speech, and generated speech. The homepage lists Universal-3.6 Pro for recordings, Universal-3.6 Pro Realtime for voice agents, and AssemblyAI-TTS-1 for text-to-speech. Model-specific language coverage matters: the listed speech recognition models support 32 languages with code-switching, while the text-to-speech model supports 6 languages.

AssemblyAI provides endpoints for converting speech into text, analyzing transcripts, and generating spoken responses. Available transcription features include language detection, formatting, word-level timestamps, custom spelling, and keyterm prompting. Speaker labeling and specialized medical vocabulary support are also available, although inclusion and additional charges depend on the selected model.

AssemblyAI exposes different APIs for different audio workflows. Recorded files use the Pre-recorded Speech-to-Text API, live audio uses the Realtime Speech-to-Text API, and short clips can use the Sync Speech-to-Text API for a finished transcript in the same request. Speech Understanding processes transcripts for summaries, sentiment, entities, and other outputs.

AssemblyAI offers managed inference through Voice AI Cloud, which the page describes as supporting unlimited concurrency for production traffic. Self-hosted deployment runs models in a customer's own VPC or on-premises for control over data. Additional endpoints include Dictation, Text-to-Speech, Guardrails, and an LLM Gateway for voice tasks.

AssemblyAI targets startups and enterprises building voice-enabled products rather than a single transcription use case. The listed solutions cover live calls, meeting documentation, human-agent assistance, clinical transcription, and spoken drafting. Healthcare workflows include medical terminology for drugs, dosages, and procedures, alongside ambient clinical notes derived from patient conversations.

AssemblyAI starts with free API access and then charges on a pay-as-you-go basis without commitments. The pricing page lists Universal-2 recorded transcription at $0.15/hr and Universal-3.5 Pro at $0.21/hr. Universal-3.6 Pro Realtime is listed at $0.45/hr, while Universal-Streaming and Universal-Streaming Multilingual are each listed at $0.15/hr.

AssemblyAI prices several capabilities as separate add-ons, including speaker diarization and Medical Mode. Some features are included with particular models but cost extra or are unsupported on others. Custom arrangements are available through sales for tailored rate limits, enhanced concurrency, and enterprise flexibility.

AssemblyAI combines recorded transcription, streaming recognition, speech generation, and transcript analysis within one API offering. Cloud and self-hosted deployment provide different infrastructure options. Model selection remains important because language support, included features, and add-on charges differ across the catalog, and some specialized modes are not yet available.