The conversational Speech to Text model
Flux is the first speech-to-text model designed for conversation, not just transcription. With built-in turn detection, ultra-low latency, and natural interruption handling, Flux enables real-time, human-like voice agents.
Integrated turn detection for natural flow
Sub-300ms end-of-turn latency
Conversational cues for agents to act on
Nova-3 level transcription accuracy
Model Overview
Deepgram models power everything from real-time conversations to domain-specific transcription, with options for speed, accuracy, and full customization.
Flux
Conversational speech recognition for real-time voice agents with built-in turn detection, natural interruption handling, and ultra-low latency.
Nova-3
High-performance speech-to-text for production transcription with top accuracy, multilingual support, and noise robustness.
Industry-tuned
Specialized speech-to-text models optimized for industry-specific vocabulary and structure for domains like healthcare, legal, and finance.
Discover Speech to Text capabilities
Deepgram’s speech-to-text features give developers everything they need to produce accurate, readable, and secure transcripts out of the box.
Keyterm prompting
Improve recognition of critical words or phrases with up to 90% higher keyword recall rate (KRR).
Learn more →
Filler words
Transcribe interruptions in speech such as “uh” and “um” to capture a more natural, human-like transcript.
Learn more →
Smart formatting
Enhance readability with automatic punctuation, capitalization, and paragraphing.
Learn more →
Power real-world solutions with Speech to Text
Deepgram’s speech-to-text API enables accurate and scalable transcription across industries, including customer support, healthcare, media, and conversational AI.
Healthcare-ready speech-to-text that captures medical terms and specialized keywords at scale. Ensure compliance with HIPAA and industry standards while reducing documentation time. Real-time transcription supports faster clinical workflows and improves patient care outcomes.
Up to 40x faster transcription creation
of pre-recorded audio than alternatives.
FAQs
A: Speech to text (STT), also called automatic speech recognition (ASR), converts spoken audio into written text. It powers use cases like transcription, analytics, accessibility, and conversational AI.
A: A speech-to-text API is a developer interface that makes speech recognition accessible in apps and services. With Deepgram’s API, you can stream audio for real-time transcription or submit recordings for batch processing at scale.
A: Yes. Deepgram can process multichannel audio to separate speakers or combine channels for clarity. Nova-3 is especially strong for meetings and call transcription.
A: Deepgram transcripts include smart formatting for punctuation, capitalization, and paragraphing, along with speaker diarization to identify who is talking. You can also enable numeral conversion so numbers are written as digits, detect filler words like “uh” and “um,” and apply vocabulary prompting to improve recognition of specialized terms. Optional redaction is available to remove personal information directly in the transcript. For a full list of transcription features, see our documentation.
A: Nova-3 is optimized for transcription at scale with best-in-class accuracy, multilingual support, and robustness in noisy environments. Flux is optimized for real-time conversation with built-in turn detection, natural interruption handling, and turn-complete transcripts.
A: 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 offers the same transcription accuracy but is optimized for real-time conversation with turn detection and low latency. For detailed benchmarks and comparisons, see our Nova-3 launch blog.
A: Deepgram’s Speech-to-Text API includes multiple models to fit your needs: Flux for real-time conversation, Nova-3 for best-in-class accuracy, Industry-Tuned models for specialized domains, Custom models trained on proprietary datasets, and Nova-2 for cost-efficient transcription. See our Models Overview for details.
A: Sign up for a free Deepgram account to access your API key. You can test models instantly in the Playground or jump into building with our starter apps on GitHub.
A: Pricing depends on the model: Nova-3 for highest accuracy, Flux for conversational AI, Nova-2 for cost efficiency, and Industry-Tuned or Custom models for specialized domains. Visit our pricing page or start free in the Playground.
A: Use Nova-3 for production transcription across meetings, media, and analytics. Industry-Tuned or Custom models may improve accuracy further for specialized domains.
A: Choose Flux for real-time conversational AI. It enables voice agents to respond naturally with turn detection, interruption handling, and low-latency events.
Trusted by startups and enterprises
Discover the power of our product through real stories.