Deepgram

Deepgram is an API platform for enterprises and developers building real-time voice AI for customer service, transcription, and conversational agents.

Key capabilities include:

• Speech-to-Text (STT) for streaming and batch transcription, designed for low latency and high accuracy on noisy audio, with options such as speaker diarization, punctuation, redaction, and keyterm prompting
• Text-to-Speech (TTS) for generating spoken responses in voice applications
• Voice Agent API that combines STT + TTS + LLM orchestration in a single real-time API, with support for bring-your-own LLM or TTS to control architecture and cost
• Audio Intelligence add-ons such as sentiment analysis, topic detection, summarization, and intent recognition

Pricing: Pay-as-you-go with $200 free credits for new users. Speech-to-text (Nova-3) starts at $0.0043/min pre-recorded and $0.0077/min streaming; text-to-speech (Aura-2) is about $0.030 per 1,000 characters; the Voice Agent API is about $0.075/min ($4.50/hour). A Growth plan (from $4,000/year) adds discounts, and Enterprise pricing is custom.

Notable for teams that need predictable performance at scale, it supports cloud or self-hosted deployments, offers custom models, and targets real-time voice agent latency (sub-300ms) while avoiding black-box agent stacks through BYO components.