In July, we unveiled our Deepgram Goes Global initiative, introducing the enterprise-grade Deepgram Dedicated deployment and our EU-hosted API to enhance regional infrastructure, compliance, and performance for large-scale voice AI operations. Now, we’re raising the bar further with four new Nova-3 monolingual models, German, Dutch, Swedish, and Danish, available for both batch (pre-recorded) and streaming (real-time) use cases. This launch represents a pivotal leap toward localized, enterprise-strength voice AI, delivering unmatched accuracy, real-time responsiveness, and operational scale across Europe’s key markets.
Enterprise Voice AI, Tuned for Europe’s Most Demanding Languages
German (de), Dutch (nl), Swedish [sv, sv-SE], and Danish (da, da-DK) each present unique challenges for speech-to-text, from compounding words in German finance, to Dutch code-switching with English in customer service, to Swedish medical terminology, to Danish phonetics that differ sharply from spelling. By bringing these languages into Nova-3, Deepgram ensures enterprises can handle these complexities with accuracy, speed, and confidence.
- Contact Centers: Deliver localized customer support that captures region-specific terms, from Dutch loyalty programs like “Bonuskaart” to Danish banking credentials such as “MitID.”
- Healthcare: Accurately transcribe specialized terminology in clinical settings, such as Swedish prescriptions like “Blodtryckssänkande läkemedel.”
- Financial Services: Improve recognition of complex terms like German “Sondertilgungsoption,” critical in loan and payment processing.
- Global Enterprises: Unify multilingual operations with consistent Nova-3 performance across Europe’s key markets.
Where Nova-3 truly stands out is Keyterm Prompting. This self-service customization tool allows enterprises to inject up to 100 domain-specific terms into the model, ensuring even the most complex or locally specific words are recognized with precision. For example, a German healthcare provider could instantly teach Nova-3 to capture drug names like “Thyroxin-Natrium” or a Swedish bank could ensure accurate recognition of “bankidentifieringskod” (bank identification code), all without retraining.
Why It Matters: The Nova-3 Advantage
In addition to Keyterm Prompting, an industry-first self-service customization tool, Nova-3 brings model-level innovations and competitive advantages that enterprises can now access in German, Dutch, Swedish, and Danish. Nova-3 doesn’t just improve on Nova-2, it extends Deepgram’s accuracy lead over competing models, especially in streaming, and introduces enterprise-ready capabilities designed for real-world performance:
- High Accuracy in Challenging Conditions: Maintains strong WER even with background noise, overlapping speech, and speaker distance, in environments like call centers, drive-thrus, and air traffic control.
- Enhanced Recognition: Accurately handles numeric sequences and specialized terminology, supporting critical domains such as banking, retail, and healthcare.
- Real-Time Performance at Scale: A unified architecture optimized for ultra-low latency makes Nova-3 ideal for mission-critical, high-concurrency deployments.
- Enterprise-Grade Adaptability: Consistent transcription quality across regional accents, technical jargon, and varied acoustic conditions.
By extending Nova-3’s enterprise-grade advantages to these four new languages, Deepgram enables organizations to deploy voice AI that is not only localized and accurate, but also fast, flexible, and resilient in real-world conditions.
Try It in Code
Switching to German, Dutch, Swedish, or Danish in Nova-3 is as simple as adding a language parameter to your request. For example, here’s how to transcribe German audio with Nova-3 using cURL:
curl --request POST \
--header "Authorization: Token YOUR_DEEPGRAM_API_KEY" \
--header "Content-Type: audio/wav" \
--data-binary @youraudio.wav \
"https://api.deepgram.com/v1/listen?model=nova-3&language=de"Replace language=de with nl, sv, or da to use Dutch, Swedish, or Danish.
Explore more options in the Models & Languages Overview.
Benchmarking: Accuracy and Recognition Gains
Nova-3 builds on the strong performance foundation of Nova-2, delivering measurable gains in both Word Error Rate (WER) and Keyword Recognition Rate (KRR) across all four new languages. While Nova-2 set a reliable standard for enterprise transcription, Nova-3 evolves that capability further, cutting more errors and improving recognition of high-value terms that enterprises depend on.
Together, these benchmarks highlight Nova-3 as a clear evolution in Deepgram’s enterprise models. It not only reduces overall errors, but also strengthens recognition of specialized, domain-specific terms in real-world applications.
Looking Ahead
This launch is the first step in a broader expansion of Nova-3’s language coverage. German, Dutch, Swedish, and Danish are only the beginning. Additional Nova-3 languages are already on the roadmap, ensuring enterprises will be able to access the same accuracy, adaptability, and enterprise-grade features across more of the world’s most widely used languages. As we extend Nova-3’s reach, our goal is clear: to provide enterprises with voice AI that is both globally accessible and locally precise.
Unlock Enterprise-Grade Voice AI Today
Sign up free and unlock $200 in credits - enough to power over 750 hours of transcription or 200 hours of speech-to-text in German, Dutch, Swedish, or Danish via our Models & Language Overview page. Dive into Nova-3’s world-class accuracy and adaptability across four new European markets.
Join a growing community of innovators: your feedback will directly influence upcoming language support, helping shape the next wave of Deepgram’s global voice AI.
Stay in the loop: Subscribe now to receive priority announcements on new features and language expansions, as Deepgram continues to scale globally with confidence.