Async is a text-to-speech API for real-time conversational products. It serves the same model that it enters in the Hugging Face TTS Arena, and it reports a median time-to-first-byte of 166 ms in its latency benchmarks. Pricing starts at $0.50 per hour of audio with a free tier.
The API offers four endpoints: a plain HTTP call that returns a full audio file, a streaming HTTP variant that starts playback before generation finishes, a timestamps variant for caption sync, and a WebSocket that accepts incremental text so an LLM's output can be voiced as it is produced. Two models are available, and drop-in integrations exist for Pipecat, LiveKit, Twilio, n8n, and Picsart Flow.
| Model | Tier | Languages |
|---|---|---|
| async_pro_v1.0 | Pro | English only, highest quality tier |
| async_flash_v1.5 | Standard | 6 languages, streaming-oriented |
Features
- HTTP synthesis: POST /text_to_speech converts a full transcript into a single audio file
- Streaming synthesis: a streaming endpoint returns audio chunks before generation completes
- WebSocket input streaming: accepts incremental text for live LLM output and voice agents
- Timestamps: a variant returns timing data for caption synchronization
- Voice cloning: clone a voice from a 3-second sample
- Languages: 15+ languages with native pronunciation handling and text normalization
- Integrations: Pipecat, LiveKit, Twilio, n8n, and Picsart Flow
- Service terms: 99.9% uptime SLA and SOC 2 compliant infrastructure on enterprise plans
