Deepgram

Deepgram is a voice AI infrastructure company founded in 2015 and headquartered in San Francisco, building speech technology for developers and enterprises. The company provides APIs for speech-to-text, text-to-speech and voice agents, using end-to-end deep learning models built specifically for audio rather than adapted from text-based systems.

Its product line includes a real-time transcription API, a text-to-speech API with latency as low as 80 milliseconds, and a Voice Agent API that combines speech recognition, speech synthesis and language model orchestration into a single interface. Deepgram serves developers building voice features directly, platform partners embedding white-label voice AI, and enterprises requiring custom models, with clients including Twilio, Cloudflare and IBM.

1. Speech-to-text API: Converts audio to text in real time for streaming and batch use cases.

2. Text-to-speech API: Synthesizes natural speech with latency as low as 80 milliseconds.

3. Voice Agent API: Combines transcription, speech synthesis and language model orchestration.

4. Audio intelligence: Extracts summaries, topics and sentiment from conversation audio.

5. Custom models: Trains domain-specific speech models for enterprise accuracy needs.

6. Developer-first design: Offers unified APIs and SDKs built specifically for audio workloads.