Speech-to-text API that's fast and accurate.
From transcription and analytics to real-time multilingual voice agents, build with accuracy, speed, and scale.
Press mic button to start transcribing
Leading voice teams build with Deepgram
Let’s talk features
Catch every word, keep up in real time, understand the context, and deploy wherever your data lives.
Real-world accuracy
Understand people clearly, across voices, accents, and real-world environments.
Ultra-low latency
Move as fast as people speak with up to 40x faster inference for real-time apps and voice agents.
Conversational context
Follow the conversation through turns, interruptions, and context, and know when it’s time to respond.
Language switching
Automatically detect languages, handle code-switching mid-utterance, and guide recognition with flexible real-time controls.
Custom vocabulary
Get your industry’s language right, from names and product terms to specialist vocabulary, with keyterm prompting and tuned models.
Flexible deployment
Run in Deepgram's cloud, route through regional endpoints, or self-host in your own environment for full control over performance, data residency, and compliance.
See it in action
Industry-leading speech-to-text performance built for production
Get the accuracy, speed, and economics you need to operate at scale.
The numbers speak for themselves
See how Deepgram performs where it counts: word recognition, inference speed, and price, with results you can check for yourself.
- >90%word recognition rate (1 - WER)
- 40xfaster inference time
- 2Xlower price
Built for different kinds of speech
From real-time voice agents to production transcription, specialized industries, and custom deployments, find the right model for the job.
Flux STT
Keep real-time conversations flowing with intelligent turn detection, natural interruption handling, and ultra-low latency.
- Real-time
- Conversational
- Low-latency
Nova-3
Engineered for production transcription with industry-leading accuracy, strong noise handling, and support for 60+ languages.
- Accurate
- Robust
- Multilingual
Industry-specific
Speak the language of your industry with models tuned for specialised terminology across medical, pharma, legal, finance, and more.
- Specialized
- Compliant
- Precise
Custom
Get better accuracy on the edge cases that matter to you by training on your own proprietary or novel datasets.
- Tailored
- Flexible
- Personalized
Related articles
Deepgram
Introducing Flux STT Multilingual: One conversational speech model for global voice agents
Announcements
Hasan Jilani
Introducing Nova-3 Medical multilingual: Medical speech recognition in 10 languages
Announcements
Deepgram
The best speech-to-text APIs in 2026
AI Trends & News
Deepgram
Automatic speech recognition vs speech-to-text: Key differences and use cases
AI Engineering & Research
FAQs
Speech-to-text (STT), also called automatic speech recognition (ASR), turns spoken audio into written text. It’s behind everything from transcription and speech analytics to accessibility and conversational AI.
A speech-to-text API lets developers add speech recognition to apps and services without building it from scratch. Deepgram’s API transcribes real-time or pre-recorded audio and powers conversational voice applications.
Use Flux STT for real-time voice agents that need conversational speech recognition and turn detection. Use Nova-3 for real-time or pre-recorded transcription. Choose industry-specific models for specialized domains like medical and pharma, or custom models for highly specialized use cases. See our speech-to-text model overview for more details.
Nova-3 delivers industry-leading accuracy with more than 50% lower word error rate (WER) compared to competitors in both streaming and batch transcription. Flux STT brings that accuracy to real-time voice agents, with turn detection and ultra-low latency. See our benchmarks page for the full comparison. See our benchmarks page for the full comparison.
Deepgram supports speech-to-text across 60+ languages and markets, with multilingual models and language controls for global applications. Available languages and features vary by model, so check the language documentation for current model and language support.
Deepgram speech-to-text pricing depends on the model, usage, and deployment. See current rates for Flux STT, Nova-3, and other speech-to-text models on our pricing page.
Create a free Deepgram account, grab your API key, and start testing models right away in the Playground.