Skip to main content
Deepgram builds voice AI models for both directions of the pipeline: speech recognition through its Nova line and speech synthesis through Aura. The company’s emphasis throughout is throughput and latency rather than maximum accuracy on difficult audio.
In Ollang Workflows, Deepgram is available as a text-to-speech provider (deepgram). See Deepgram Aura Text-to-Speech. This page documents the speech recognition side for reference; Ollang’s transcription providers are listed in the Speech-to-Text catalog.

​
Model and coverage

Deepgram’s advantage is throughput and latency rather than accuracy on mixed-language audio, where AssemblyAI’s Universal-3.5 Pro and ElevenLabs Scribe v2 are the stronger documented options.

​
Key capabilities

  • Real-time processing with low latency, built for live captioning and voice agents.
  • Custom model training on domain vocabularies where standard models under-perform.
  • Flexible deployment across cloud, on-premises, and hybrid, for privacy and residency requirements.
  • GPU-based inference enabling high parallelism relative to CPU-bound alternatives.

​
Reference