Deepgram
Voice AI APIs for speech-to-text, text-to-speech, and real-time voice agents
About Deepgram
Deepgram is a developer platform for voice AI. Its Nova-3 and Flux models transcribe streaming and pre-recorded audio through REST and WebSocket APIs, and the same account covers Aura and Flux text-to-speech, a Voice Agent API, and audio intelligence such as summarization and sentiment. Billing is per second of audio, with a $200 free credit to start and self-hosted deployment for enterprises.
Pros & Cons
Pros
- Low per-minute price: Nova-3 batch transcription costs about $0.26 per hour on pay as you go, with true per-second billing and no rounding
- $200 free credit with no credit card lets you test at real volume before paying
- Fast streaming transcription over WebSocket, and Flux adds turn detection built for voice agents
- One account covers STT, TTS, a Voice Agent API, and audio intelligence
- Clear developer docs, SDKs, and an API playground make integration quick
- Enterprise options include self-hosting, an EU endpoint, SOC 2, HIPAA BAA, and GDPR
Cons
- Add-ons such as redaction, keyterm prompting, entity detection, and streaming diarization are billed per minute on top of the base rate
- Several streaming prices are limited-time promotional rates, so long-term costs can rise to the regular price
- Flux doesn't support speaker diarization or smart formatting, so voice agents that need speaker labels must use Nova-3
- Accuracy drops with heavy accents, overlapping speech, and some non-English languages, per user reviews
- It's an API, not an end-user app; you need developer work to get transcripts
Key Features
Nova-3 speech-to-text
General-purpose transcription model for streaming and pre-recorded audio, covering 45+ languages across Nova models, with a multilingual mode that handles code-switching mid-sentence.
Flux conversational STT
Streaming model built for voice agents with turn-based transcription and Start of Turn, End of Turn, and Eager End of Turn events, available in English and a multilingual variant.
Speaker diarization
Labels who spoke when in multi-speaker audio. Included at no extra cost on pre-recorded audio; a paid add-on for streaming. Supported on Nova-3, not on Flux.
Word-level timestamps
Returns start and end times for every word, on both Nova-3 and Flux, for captions, search, and audio alignment.
Smart formatting and keyterm prompting
Smart formatting adds punctuation, casing, dates, and currency at no charge; keyterm prompting boosts recognition of product names, jargon, and acronyms.
Redaction and entity detection
Removes PII such as card numbers, SSNs, and phone numbers, and extracts names, organizations, locations, and dates from transcripts.
Text-to-speech (Aura and Flux TTS)
Low-latency TTS voices for conversational apps, billed per 1,000 characters, with Aura-1, Aura-2, and Flux TTS models.
Voice Agent API
Single WebSocket API that combines STT, an LLM, and TTS to run real-time voice agents that handle interruptions and call functions, with bring-your-own LLM and TTS options.
Audio intelligence
Summarization, topic detection, sentiment analysis, and intent recognition over transcripts and text.
Self-hosted and EU deployment
Enterprise customers can run models on-premise or in a private cloud, and a dedicated EU endpoint keeps processing inside the European Union.
Pricing
- $200 free credit, no credit card required
- No minimums, credits don't expire
- Nova-3 pre-recorded: $0.0043/min (about $0.26/hr)
- Nova-3 streaming: $0.0048/min (promo; regular $0.0077/min)
- Flux English streaming: $0.0065/min (promo)
- Aura-2 TTS: $0.030 per 1K characters
- Voice Agent API Standard: $0.075/min
- Up to 150 concurrent streaming connections
- Per-second billing
- Pre-paid annual credits, save up to 20%
- Nova-3 pre-recorded: $0.0036/min
- Nova-3 streaming: $0.0042/min (promo)
- Aura-2 TTS: $0.027 per 1K characters
- Voice Agent API Standard: $0.068/min
- Up to 225 concurrent streaming connections
- Custom-trained models
- Volume discounts
- On-premise or private cloud deployment
- HIPAA BAA available
- Dedicated support and SLAs
Best For
Voice agents and IVR
Use Flux streaming STT with end-of-turn detection, or the Voice Agent API, to build phone and web agents that respond with low latency and handle interruptions.
Contact center analytics
Transcribe recorded calls in batch with diarization, redaction, and sentiment to score agents and spot trends at a low per-hour cost.
Meeting and podcast transcription
Process pre-recorded audio with Nova-3, speaker labels, and word timestamps to produce searchable transcripts and captions.
Live captioning
Stream audio over WebSocket to caption events, broadcasts, and video calls in real time.
Medical and regulated transcription
Use self-hosted deployment, HIPAA BAAs, and PII redaction to transcribe sensitive audio within compliance requirements.
Similar Tools
ElevenLabs
AI voice generator and voice agents platform
Murf AI
AI voice generator with 200+ realistic text-to-speech voices
TTSOpenAI
Advanced AI voice engine for natural text-to-speech
LOVO AI
AI voice generator and video editor with 500+ voices in 100+ languages
Groq
Ultra-fast AI inference powered by custom LPU silicon
EKHOS AI
Secure offline AI transcription for legal, medical, and sensitive recordings
Featured In
Best Speech-to-Text APIs for Developers Building Voice Apps (2026)
Best for real-time voice agents where turn detection and streaming latency matter more than language count.
Best Real-Time Text-to-Speech APIs for Game Characters and AI Companions (2026)
Best for teams that want speech-to-text and TTS from one vendor for grounded companion or tutor characters, rather than heavily acted NPCs.