customers
All customer stories
Top Voice AI companies are building with Assembly.
resources
Latest Release
Text-to-Speech API
Our first text-to-speech model, AssemblyAI-TTS-1, is built for voice agents, on live calls. It sounds natural, responds fast, and keeps callers on the line.
resources
Stream audio over a WebSocket and get formatted, immutable transcripts back at ~150 ms p50 from the Realtime Speech-to-Text API—accurate enough for production voice agents. Open a connection and start transcribing in a few lines of Python or JavaScript.
Hey, thanks for calling—can I grab your account number?
Sure, it's 4471-208-93—let me know if you need the zip too.
Got it, pulling that up right now.
Universal-3.6 Pro Realtime returns partial and final transcripts at ~150 ms p50—formatted, immutable, and fast enough to drive live voice agents and agent assist.
Native turn detection, speaker labels, and keyterm prompting handle the hard parts of streaming audio out of the box.
Universal-3.6 Pro Realtime streams 32 languages with automatic language detection and code-switching. Universal Streaming Multilingual covers English, Spanish, French, German, Italian, and Portuguese at $0.15/hr.
36%
improvement in close rate
“The new Universal-3.5 Pro speech model from AssemblyAI is best so far in terms of accuracy, latency, and language switching.”
80%
increase in customer satisfaction
“Assembly has saved us countless hours managing models, and provided exceptional accuracy.”
36%
improvement in close rate
“The new Universal-3.5 Pro speech model from AssemblyAI is best so far in terms of accuracy, latency, and language switching.”
80%
increase in customer satisfaction
“Assembly has saved us countless hours managing models, and provided exceptional accuracy.”
Grab your API key and connect to the streaming WebSocket with the Python or JavaScript SDK—no infrastructure to stand up. Handle partial and finalized turns with a single event handler and start building. Universal-3.6 Pro Realtime is $0.45/hr and Universal Streaming is $0.15/hr, billed on session duration.
Start building freeNo credit card required
Streaming word error rate measures realtime Speech-to-Text accuracy against a human reference—the core signal for anything that acts on live audio.
Normalized WER on the voice-agent caller corpus, mean of three noise conditions.
*Lower is better*Source: AssemblyAI realtime benchmarks technical report v2026.09 (assemblyai.com/benchmarks), data 18 Aug–24 Sep 2026.
