Deepgram Aura-2

Deepgram Aura-2 is Deepgram's enterprise-grade text-to-speech model launched in 2025, purpose-built for high-volume real-time voice agent pipelines. It achieves sub-200ms baseline time-to-first-byte latency with an optimized path reaching 90ms, delivering a Real-Time Factor of approximately 0.111x, meaning one second of audio is synthesized in around 100 milliseconds. It is designed as the TTS layer of a bring-your-own-STT and bring-your-own-LLM cascade, integrating directly with Deepgram's Nova-3 ASR or any external STT provider via the Deepgram Voice Agent API or standalone HTTP and WebSocket endpoints. Aura-2 is billed at approximately $0.0150 per 1,000 characters of TTS output, translating to around $0.011 per minute at typical agent speech rates. It ships with 20-plus preset voices and supports English, Dutch, French, German, Italian, and Japanese following language expansion in 2025. The model is accessible via Deepgram's Python, JavaScript, Go, Ruby, and .NET SDKs, with WebSocket streaming for real-time output and telephony platform connectors for Twilio, Vonage, and similar providers. Deepgram Aura-2 is the professional-tier TTS choice for high-volume telephony operators who need the lowest combined latency and cost across thousands of simultaneous AI phone calls. Its sub-90ms latency outperforms most cascade TTS components, and its per-character pricing makes it economical at scale. Teams requiring a full managed voice agent platform can pair it with Deepgram's Nova-3 STT and Voice Agent API; teams wanting full control can use Aura-2 as a standalone TTS layer with any LLM and STT of their choice.