AssemblyAI has released AssemblyAI-TTS-1, its first text-to-speech model, now generally available. The AssemblyAI-TTS-1 text-to-speech API is designed for voice agents operating on live phone calls—a setting where latency and pronunciation accuracy often determine whether a caller stays on the line. Pricing is $15 per 1M characters, with no annual contract and no concurrency tier to purchase. For developers already using AssemblyAI's speech-to-text, the new model closes the voice loop with a single vendor.
TL;DR
- Deployment: Hosted API model.
- Availability: AssemblyAI's cloud only, in US (us-east-1) or EU (eu-north-1). No open weights or on-prem option. Any client supporting HTTP or WebSocket works.
- Performance: Ties ElevenLabs v4 on entity accuracy in AssemblyAI's internal tests at under one-fifth of its price.
- Best case: 47 ms median time to first byte across six models in AssemblyAI's API test (self-reported).
- Worst case: 159 ms median streaming latency on Cekura's independent benchmark—8th among 19 timed models.
- Bottom line:
- Best: $15 per 1M characters with unlimited concurrency.
- Worst: No SSML, speed control, pronunciation dictionaries, or emotion tags yet.
via MarkTechPost