What is Deepgram ASR?
Deepgram ASR is a voice AI service for developers and organizations building applications that listen to speech and generate spoken responses. Deepgram ASR combines speech-to-text, text-to-speech, voice agents, and audio intelligence APIs. The service covers both streaming and pre-recorded transcription, alongside conversational workflows that connect speech recognition, language model orchestration, business logic, external systems, and speech generation.
What Deepgram ASR can do
Deepgram ASR provides speech recognition models and optional processing features for turning audio into structured, usable text. Nova models support 60+ languages, and Nova multilingual models handle multiple languages in real time. Deepgram ASR also supports generating speech from text and analyzing transcriptions, allowing developers to combine transcription, spoken responses, and conversation analysis within voice applications.
- Deepgram ASR transcribes streaming and pre-recorded audio through APIs.
- Deepgram ASR distinguishes speakers with speaker diarization.
- Deepgram ASR offers Smart Formatting, Keyterm Prompting, redaction, and entity detection.
- Deepgram ASR generates speech through batch requests and WebSocket streaming.
- Deepgram ASR supports appointment-booking voice agents through function calling.
- Deepgram ASR provides summarization, topic detection, sentiment analysis, and intent recognition.
How Deepgram ASR works
Deepgram ASR exposes APIs that developers connect to their applications. The website describes sending chatbot messages to Flux TTS over WebSocket and playing the resulting audio in a browser, as well as streaming LLM replies into speech from a Python application. The Voice Agent API combines speech-to-text, a managed language model, and text-to-speech, with options to bring an external language model or TTS provider.
Deepgram ASR offers account signup, documentation, demo requests, and community resources including GitHub and Discord. Cloud processing includes a dedicated EU endpoint for European Union data residency. Enterprise customers with requirements that prevent cloud processing can use self-hosted containers in their own VPC or on-premises hardware. These self-hosted deployments require NVIDIA GPUs.
Who Deepgram ASR is for
Deepgram ASR serves developers building voice applications and enterprise teams with higher volumes, deployment requirements, or support needs. The supplied examples focus on conversational applications: adding speech to a chatbot, streaming language model responses, comparing generated voices, and booking appointments through a receptionist. The website also discusses contact center environments and voice interactions that span multi-turn conversations.
- Developers can add spoken responses to existing chatbots.
- Voice application teams can stream language model replies as audio.
- Businesses can build receptionists that confirm details before booking appointments.
- Teams can extract summaries, topics, sentiment, and intent from transcriptions.
Deepgram ASR pricing and plans
Deepgram ASR starts with $200 in free credits that do not expire, and no credit card is required to begin. Pay as you go has no minimum commitment. Growth starts at $4,000+ / year in prepaid credits redeemed against actual usage, with savings up to 20%. Enterprise uses custom pricing for larger volumes, deployment requirements, and support needs.
Deepgram ASR bills speech-to-text per second of processed audio without block-based rounding for normal requests. Text-to-speech is billed by input characters, Voice Agent usage by WebSocket connection time, and Audio Intelligence by input and output tokens in addition to transcription. Smart Formatting is included at no cost, while several other transcription features carry additional charges.
Deepgram ASR strengths and limitations
Deepgram ASR brings speech recognition, speech generation, and voice agents together, with both cloud and Enterprise self-hosted deployment options. Non-expiring starter credits support prototyping without a credit expiration deadline. Published billing rules distinguish transcription, synthesis, connection time, and analysis costs, while Enterprise options address data sovereignty and security requirements. Usage limits and feature-specific charges remain important considerations when estimating application costs.
- Concurrency limits vary by plan and region, and excess requests can be queued or rejected.
- Audio Intelligence and optional transcription add-ons increase costs beyond base transcription.
- Self-hosted deployments require Enterprise access and NVIDIA GPUs.
- Growth overages are billed at the Growth rate plus 10% unless customers renew early.
Deepgram ASR FAQ
What is Deepgram ASR?
Deepgram ASR is a voice AI service with speech-to-text, text-to-speech, voice agent, and audio intelligence APIs. Deepgram ASR supports streaming and pre-recorded transcription for developers and enterprise teams building conversational applications and audio-processing workflows.
Is Deepgram ASR free?
Deepgram ASR offers $200 in free credits that do not expire, with no credit card required to start. Subsequent usage is billed through Pay as you go, prepaid Growth credits, or a custom Enterprise plan rather than an unlimited free tier.
What languages does Deepgram ASR support?
Deepgram ASR provides Nova speech recognition models supporting 60+ languages. Nova multilingual models support multiple languages in real time, and the pricing page lists both English and multilingual Flux STT options.
Can Deepgram ASR be self-hosted?
Deepgram ASR supports self-hosted containers on Enterprise plans, running in a customer's own VPC or on-premises hardware. Self-hosting requires NVIDIA GPUs and is offered for teams whose data sovereignty or security requirements rule out cloud processing.
Does Deepgram ASR charge extra for transcription features?
Deepgram ASR includes Smart Formatting at no additional cost, but redaction, Keyterm Prompting, entity detection, and speaker diarization have separate add-on rates. Audio Intelligence features are also billed separately by input and output tokens on top of transcription.
Can Deepgram ASR power a voice chatbot?
Deepgram ASR supports spoken chatbot replies through Flux TTS over WebSocket and provides a Voice Agent API combining speech recognition, a managed language model, and speech generation. Developers can also bring their own language model or TTS provider for a lower per-minute rate.
From the community
No reviews yet
Be the first to review this tool. What you write helps the next person choose.
