Skip to main content
Deepgram Aura is Deepgram’s text-to-speech line, engineered for low-latency, high-throughput voice. Aura-2 is the current generation, aimed at enterprise applications — customer service, healthcare, sales, interviews, and IVR — where clarity and consistency matter more than expressive range. Its voice catalog is unusual in shape: deep rather than wide. Seven languages, but more voices inside English and Spanish than any other engine in this catalog offers.
Available in Ollang Workflows as deepgram — AI Dubbing. See the Text-to-Speech catalog. Deepgram’s speech recognition side is documented separately under Deepgram.

Voices and languages

Aura-2 provides 88 voices across 7 languages: Voices are addressed as [model]-[voice]-[language], for example aura-2-thalia-en. Aura 1 remains available as a legacy generation with 12 English voices.

Key capabilities

  • Accent depth within languages — five English accents and four Spanish accents mean regional variants can be cast properly rather than approximated. For an English-to-Spanish program targeting several Latin American markets, this is a genuine advantage over engines with one Spanish voice.
  • Range of speaking styles across genders, age groups, and delivery — warm conversational, confident professional, and characterful storytelling.
  • Low latency at scale — the architecture is built for real-time voice agents, which translates into fast batch throughput for dubbing.
  • Efficient batch throughput, which suits large instructional and corporate libraries.

Where it fits in a workflow

Deepgram is a good fit when your target languages sit inside its seven and the content is instructional, corporate, or informational. For an English-to-Spanish training library the accent depth is a real product advantage rather than a rounding difference.

Trade-offs

  • Seven languages is the narrowest coverage in the catalog. Confirm your language pair before selecting it — this is the most common reason a Deepgram Workflow has to be reconfigured. French in particular has only two voices.
  • Built for voice agents, not narration. On emotionally driven or performed content it reads accurately but does not act.
  • No voice cloning.

Reference