AssemblyAI launches TTS-1 for voice agents: 90ms latency, 4,000+ voices at $15 per 1M characters
AssemblyAI · x · 2026-10-28
AssemblyAI announced its first text-to-speech model, AssemblyAI-TTS-1, built specifically for voice agents on live calls.
Key specs:
- 90 ms time to first audio, aimed at natural real-time conversation
- Trained on millions of hours of audio; claims leading accuracy on independent benchmarks
- Instant voice cloning plus a library of 4,000+ voices, with new ones added daily
- Priced at $15 per 1M characters, covering all supported languages and regions
The release rounds out AssemblyAI's platform spanning transcription, speech understanding, and now voice interaction, with a demo showing a kid ordering pizza by voice.
Related event: AssemblyAI Launches First TTS Model TTS-1(3 posts)→
More from Multimodal
- Musk's Grok-generated AI Odyssey panned by Guardian as "dull as a laptop screensaver" — nordicinst · 2026-10-29
- Artificial Analysis launches image-to-video benchmark: Seedance 2.5 tops, MiniMax H3 best value — ArtificialAnlys · 2026-10-29
- Baidu's open OCR model digitizes all 270 pages of an 1899 Merck manual on one GPU — MaziyarPanahi · 2026-10-29
- Baidu Open-Sources Unlimited OCR 2: 270-Page 1899 Drug Manual Parsed in One Request — MaziyarPanahi · 2026-10-29
- Sunra launches API aggregating leading image, video, audio and 3D models at maker prices — HeyToha · 2026-10-29
- Side-by-side videos show the leap from generation 1 to generation 14 of video models — JohnPhamous · 2026-10-29