Aura-2
Deepgram's multilingual TTS for the enterprise voice stack: 7 languages, 90 voices, streaming over WebSocket — the one-shot counterpart to its Flux agent model.
About
Aura-2 is Deepgram's broadest-language text-to-speech model, built for enterprises that want STT and TTS from one provider inside the same low-latency infrastructure. It serves English, Spanish, German, French, Dutch, Italian, and Japanese through the /v1/speak endpoint, with 90 catalog voices and WebSocket streaming for real-time use (the model name concatenates with the voice, e.g. aura-2-thalia-en). At launch Deepgram claimed preference-testing wins over ElevenLabs, Cartesia, and OpenAI for conversational enterprise use cases — but never published the methodology, so treat that as marketing. Since August 2026 Deepgram has pointed English-first voice-agent builders at Flux TTS instead; Aura-2 remains the pick when you need a language Flux doesn't cover, or one-shot synthesis rather than barge-in-capable agent streaming. Max input is 2,000 characters per request, and unlike some rivals Aura-2 does not do voice cloning — it's a fixed voice catalog.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Provider-reported preference testing vs ElevenLabs, Cartesia, OpenAI at launch (no methodology published)
- No independent benchmarks found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (TIER metrics, metrics-v1). Evidence: Deepgram's enterprise voice platform (200k+ developers, Fortune 500 customers); public skills repo and docs; now complemented by Flux TTS. Strengths: one-provider STT+TTS stack, 90 voices across 7 languages, WebSocket streaming, enterprise compliance posture. Docked for: (1) closed weights and the launch preference-testing claim came with no published methodology; (2) pricing is not on a clean public page — the $30/1M-char figure is third-party-quoted; (3) no voice cloning, and Deepgram now steers new English agent builds to Flux, so Aura-2's long-term positioning is narrowing. License: proprietary.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Deepgram's multilingual TTS: 7 languages (EN, ES, DE, FR, NL, IT, JA), 90 voices, /v1/speak + WebSocket streaming
- · Max 2,000 chars/request; no voice cloning (fixed voice catalog)
- · Launch preference-testing claim vs ElevenLabs/Cartesia/OpenAI came with no published methodology
- · Deepgram now steers new English agent builds to Flux TTS; Aura-2 remains the multilingual/one-shot option
- · Pricing opaque on public page — $30/1M chars is third-party-quoted