AssemblyAI Launches TTS-1: $15 Per Million Characters, Unlimited Concurrency, No Contracts Required
AssemblyAI has rolled out TTS-1, its first text-to-speech model purpose-built for voice agents. Priced at $15 per million characters, the service promises unlimited concurrency with no mandatory contracts, and supports 6 languages at launch. Whether its low price and concurrency guarantee hold up for real-world calls depends on its latency, audio quality and operational stability.
AssemblyAI has released TTS-1, its very first in-house text-to-speech model, with a clear target use case: powering voice agents.
As laid out clearly by MarkTechPost author Michal Sutter, the key details are straightforward: $15 per million characters, unlimited concurrency, no contracts required, with support for 6 languages at launch. The original announcement does not specify exactly which 6 languages are included, so developers need to check official documentation before integration.
This company was long primarily recognized as a speech-to-text vendor. By filling in the text-to-speech product line now, voice agents can handle both the "listening" and "speaking" functions via APIs from the same provider. For developers, this means one less vendor to integrate, which simplifies both billing workflows and troubleshooting processes.
TTS for Voice Agents: Cost Structure Draws More Attention Than Timbre at First
The standard voice agent pipeline is not complicated: ASR handles listening, LLM takes charge of reasoning, and TTS delivers spoken output. TTS is billed per character, so its unit price directly factors into the cost of every call.
Based on a rough estimate of 900 to 1000 characters per minute for English, 1 million characters equals roughly 16 to 18 hours of audio, though the exact figure varies based on speaking speed, pauses and source text content. At this price point for the voice agent scenario, teams have good reason to recalculate their unit economic models.
Unlimited concurrency eliminates queuing issues, but it does not improve raw response speed. The real pain points for voice agents include first-utterance latency, interruption handling capability, and long conversation stability. None of these metrics are mentioned in the currently public information for TTS-1.
Key Factors Developers Should Evaluate
Do not only focus on the unit price. Run tests with your real call scripts, and measure three things: first-packet latency, continuous synthesis stability, and accent performance for your target market across the 6 supported languages.
The no-contract policy and unlimited concurrency lower the barrier for trial and error, but a low unit price does not equal a low total cost. ASR, LLM, orchestration, and telephony lines are still all part of your final bill.
If you are building voice agent products, TTS-1 is well worth adding to your candidate list. Test it with your actual production text first, before planning large-scale deployment.
Source: MarkTechPost
发布时间: 2026-10-29 15:28