Deepgram Aura 2
Aura 2 baseline — Coval: 327ms median TTFA, 5.1% WER (12 Aug 2026)
Comparative Scores
Architecture
Relevant for Phase 1 MVP if using Deepgram Nova-3 for ASR. The ASR+TTS stack from a single provider simplifies integration and reduces latency. Limited to English and no voice cloning are significant constraints.
Analysis
Deepgram Aura 2 remains the earlier TTS option in the Deepgram family, optimized for English voice-agent pipelines. The vendor historically claimed <100ms TTFA; the current independent Coval/Openbenchmarks baseline is 327ms median TTFA, 547ms p95 and 5.1% WER for aura-2-thalia-en (12 August 2026). Use it as a measured Deepgram baseline, not as evidence for the new Flux TTS model, whose latency and WER are currently vendor-reported only. No voice cloning, limited expressiveness — focused on predictable English voice-agent operation.
Strengths
- Independent baseline: 5.1% WER (Coval, 12 Aug 2026)
- Optimized for voice agent pipelines
- Natural pairing with Deepgram ASR
- Reliable at scale
Weaknesses
- No voice cloning
- English only (Aura 2)
- Limited expressiveness
- No lip-sync data
Voice Capabilities
No voice cloning. Pre-built voices only.
Aura provides clean pace and pronunciation control, but no documented programmable emotion.
How: `speed` (0.7–1.5) and IPA pronunciation dictionary.
Validate: For emotional scenes, plan distinct writing and an expressive engine rather than attribute control it does not expose.
Vendor claim: <100ms TTFA. Independent Coval/Openbenchmarks measurement (12 Aug 2026): 327ms median TTFA, 547ms p95 and 5.1% WER for Aura-2-Thalia-en. Often paired with Deepgram Nova-3 ASR for a full stack.
No native lip-sync data.
Pricing
$15/1M chars. Often bundled with Deepgram ASR for full voice agent stack.
| Plan | Subscription/mo | Included | Overage/min | Max streams | Top-up |
|---|---|---|---|---|---|
Pay As You Go Same plan as Deepgram STT | Free | $200 free credit | $0.0150 | 150 | Yes |
GrowthRecommended Volume discount available | $333/mo | Prepaid credits, up to 20% discount | $0.0120 | 225 | Yes |
Sovereignty & Compliance
Cloud only. Enterprise on-premise via agreement.
Data residency: US, EU
Deepgram Aura 2 — Strategic Positioning
Beyond technical specs: where does this tool sit in the ecosystem, what are the risks and strategic implications for GamiWays?
Deepgram Aura is the unified voice platform play: STT+TTS+LLM in one API, sub-200ms latency, $1.3B valuation — but its VPC-only stance limits sovereignty appeal for regulated European deployments.
A. Strategic Positioning
Target customer: Developer / Enterprise — real-time voice agents, unified STT+TTS platform
Sub-200ms TTS latency as part of a unified STT+TTS+LLM platform — the one-stop shop for voice AI agent infrastructure.
B. Competitive Moat
- Unified STT+TTS+LLM API — reduces integration complexity and latency vs multi-vendor stacks
- Sub-200ms TTS latency with natural voices — competitive with Cartesia for real-time agents
- Series C $130M (Jan 2026) — $1.3B valuation — financial strength for R&D
Vulnerability: No on-premise TTS option (VPC only). Open-source models catching up. Pricing pressure from competitors.
E. Strategic Questions for GamiWays
Sovereignty fit
EU data residency via VPC available. No full on-premise for TTS. Moderate sovereignty fit — better than pure cloud, worse than Inworld.
Build vs. Buy
Buy for Phase 1 (unified platform, low integration overhead). Evaluate on-premise alternatives for Phase 2 sovereignty.
Lock-in risk
Unified STT+TTS platform creates integration lock-in. VPC deployment and competitive pricing reduce dependency risk.
Roadmap alignment
Good for Phase 1 (unified STT+TTS for voice agents). Phase 2 requires on-premise TTS — consider Inworld or open-source for sovereignty.
This sheet's verification
Coval/Openbenchmarks — Aura-2-Thalia-en, 12 Aug 2026 (327ms median TTFA, 547ms p95, 5.1% WER)
Update note: Mise à jour Flux TTS : baseline Aura-2 indépendante ajoutée depuis Coval/Openbenchmarks. Les mesures Aura ne sont pas attribuables à Flux TTS.
This date covers provider-specific prices, capabilities and notes. Comparative benchmarks follow the synchronization and methodology shown above.