Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
TL;DR: In a natural voice agent conversation, anything over 500ms end-to-end feels stilted. Transcription latency is the first line item in that budget, and it sets the ceiling for everything downstream. This guide walks through wiring Solaria-1 into a Pipecat pipeline, tuning VAD thresholds, and handling failure modes, so the STT layer is a decision you can revisit without a rewrite. Because each component is independently swappable, choosing your STT provider is an architectural constraint you control, not one the framework imposes.
TL;DR: Voice agent latency comes from every layer, but the LLM accounts for the largest share of your total budget. Self-hosting open-source STT models burns GPU budget on cold-start delays and accuracy that degrades on accented speech. Integrating our Solaria-1 streaming API with LiveKit gives you partial transcripts in under 103ms and final transcripts at approximately 300ms, with true code-switching across 100+ languages. This guide delivers production-ready Python and Node.js code to connect LiveKit's audio egress to Gladia, configure silence detection, and drive natural turn-taking using live partials.
TL;DR: Recall.ai handles the platform-level complexity of joining Zoom, Meet, and Teams calls while we handle transcription, diarization, and enrichment. This guide walks through the full integration: spawning a bot, routing recorded audio to our async API, enabling pyannoteAI Precision-2 diarization, and mapping speaker labels to participant names using timestamp overlap. Choose Solaria-3 for post-meeting accuracy on English and European business audio, Solaria-1 for real-time captions or broad language coverage. On Growth and Enterprise plans, your audio is never used to train our models.
Introducing Solaria-1, the first truly universal speech-to-text model
Published on April 2, 2025
Voice is the most natural way we communicate. As AI continues to redefine the way businesses interact with customers, the ability to accurately and instantly transcribe speech across languages is no longer a luxury—it’s a necessity. Enter Solaria-1, the breakthrough speech-to-text model designed to power the next era of global AI-driven conversations.
Whether you’re developing industry-specific voice agents or delivering high-performance customer experiences, Solaria-1 provides the foundation you need —withunmatched language coveragewithout compromising on quality or speed.
With best-in-class real-time transcription at an industry-leading 94% WAR (Word Accuracy Rate) in English and other common languages, exclusive support of 100+ languages, 42 of which are unique to Gladia, and ultra-low 270 ms latency, Solaria-1 is setting the new standard for AI-driven voice interactions.
Update: new model released
Since publishing this article, Gladia has released Solaria-3 — our newest speech model, built specifically for real-world business audio: noisy, fast-paced, and conversational. On production recordings, Solaria-3 ranks #1 across English and core European languages (EN, FR, DE, ES, IT), beating AssemblyAI, ElevenLabs, Deepgram, Mistral, and Speechmatics. It’s also 26% more accurate than Solaria-1 on real English customer calls. That said, the two models are built to complement each other, not compete. Solaria-1 remains the better choice if you need broad language coverage (100+ languages), code-switching support, real-time streaming, or if your audio is clean, formal, or institutional, such as parliamentary recordings. Solaria-3 is the upgrade if your priority is accuracy on European business audio, call center recordings, or anything noisy and conversational. Not sure which to use?
In industries where every word carries weight—whether it’s customer support, financial transactions, or legal transcriptions—precision is key. Solaria-1 is engineered to capture even the most nuanced speech patterns with human-level accuracy, making it the go-to model for enterprises that demand nothing less than excellence.
We deliver industry-leading accuracy of transcription across English, Spanish, and other widely spoken languages while outperforming competitors in complex scenarios such as noisy environments, accented speech, and domain-specific terminology.
To ensure results reflect real-world performance, we benchmarked Solaria-1 using public datasets like Mozilla’s Common Voice and Google’s FLEURS—both designed to challenge STT models with diverse accents, dialects, and audio conditions.
While many providers test only on Common Voice version 16, we evaluated across multiple versions to avoid tuning our model to a single release. We also ran tests using anonymized enterprise datasets from contact center environments, measuring Solaria-1 against each competitor’s most advanced model—beating Deepgram’s English-only Nova-3 with our multilingual model.
Accuracy
Average Word Accuracy Rate (WAR) in English. Higher is better.
Solaria94%
Deepgram93.5%
AssemblyAI91.5%
Latency
On Common Voice. Final complete processing time. Lower is better.
•••
Solaria
698 ms· 103 ms on interrupt
Deepgram1158 ms
AssemblyAI1278 ms
Low-latency for seamless communication
In voice AI, latency doesn’t just impact performance—it defines the experience. Whether you're building real-time voice assistants or multilingual support bots, responsiveness is what makes interactions feel natural. Solaria-1 delivers ultra-low latency across the board, enabling smooth, human-like conversations.
To understand how we measure latency and ensure our API delivers the best performance in real-life, enterprise environments, it's useful to make a distinction between two key metrics: latency on interrupt and latency on final, which we break down below.
Latency on Interrupt (aka 'Time to First Byte')
When a user starts speaking, how long does it take for the AI to begin responding? This is known as latency on interrupt, or Time to First Byte (TTFB)—and it’s one of the most important benchmarks in voice AI. When companies claim their voice tech is “more responsive” or “more natural,” this is the metric they’re pointing to.
5002500
103 ms
202 ms
465 ms
Solaria-1DeepgramAssemblyAI
Imagine speaking to a voice assistant like Siri and interrupting mid-sentence. The delay before the system reacts? That’s interruption latency. The faster the response, the more human the interaction feels.
With an average response time of 270 milliseconds, Solaria-1 positions itself among the most responsive speech-to-text models available today—delivering fluid, real-time interactions that feel intuitive and immediate, while making sure to strike the right balance between latency and accuracy of transcription across languages.
Latency on Final
Equally important is latency on final—how quickly the system delivers a complete transcript once the user finishes speaking. This determines when your downstream AI (like an LLM) can start processing the request.
Solaria-1 delivers final transcripts in just ~698ms, outperforming competitors by more than half a second. That speed can significantly accelerate AI response times, making your overall system feel faster, smarter, and more responsive.
140010507003500
698 ms
1158 ms
1278 ms
Solaria-1DeepgramAssemblyAI
Latency on Final among leading STT providers (lower is better)
We speak the languages they don’t
Expanding globally means more than just supporting a handful of dominant languages. It requires an AI model that understands the full spectrum of linguistic diversity and can deliver native-level recognition no matter where it’s deployed.
Solaria-1 is the only speech AI model offering native-level accuracy across 100 languages, including 42 that are completely unsupported by competitors. This includes widely spoken but underserved languages such as:
Critical business regions: Hebrew, Pashto, Kazakh, Georgian, Mongolian.
Emerging voice AI frontiers: Haitian Creole, Maori, Javanese, Malagasy.
Being truly multilingual means more than just transcription. Solaria-1 enables real-time code-switching, allowing users to shift naturally between languages—a must for global customer interactions. It also supports real-time translation across all supported languages, helping teams eliminate communication barriers and connect with users anywhere.
Enterprise-grade adaptability & customization
Precision in speech AI isn’t just about general accuracy—it’s about being accurate where it matters the most for your business. Solaria-1 delivers best-in-class custom vocabulary andnamed entity recognition (NER) for real-time applications, allowing platforms to:
Custom vocabulary training From medical diagnoses to financial terms, Solaria-1 can be trained to recognize industry-specific jargon, ensuring specialized language is captured with precision.
Brand and product name adaptation Unlike generic speech models, Solaria-1 recognizes and accurately transcribes company names, product terms, and acronyms, reducing transcription errors in brand-sensitive environments.
Key data extraction The model can identify and extract phone numbers, email addresses, and postal addresses, allowing businesses to automate and streamline workflows in a way that traditional speech recognition tools cannot.
Fine-tuned language sensitivity Solaria-1 minimizes false positives and misinterpretations by adjusting sensitivity settings per language, ensuring that technical terms, slang, and colloquialisms are properly understood.
Whether handling legal transcriptions, technical support calls, or AI-driven sales conversations, Solaria-1 ensures every interaction is smarter, faster, and more precise.
Designed for global scalability
Large-scale voice AI applications require infrastructure that’s not just powerful, but scalable and reliable. Solaria-1 is built to handle enterprise-level deployments with ease, offering:
Robust multi-region infrastructure: Built for stability, with dedicated deployment options in both US & EU-based regions to meet diverse operational needs.
Future-proof AI: Constantly evolving, Solaria-1 is designed to integrate seamlessly with next-generation voice technologies, keeping businesses ahead of the curve.
Enterprise-grade data security: In time for the release, we're now fully compliant with GDPR, HIPAA, and SOC 2.
New partnerships
As part of the Solaria-1 launch, we’re excited to announce partnerships with two leading developer frameworks in the voice agent space: LiveKit and Daily (the team behind Pipecat).
LiveKit is now using our live translation capabilities in AI-driven applications, while Daily has built a demo chatbot powered by Gladia that can switch languages on the fly—try it out for fun! Our API is now natively integrated with both libraries, all powered by Solaria-1.
Unlock the next frontier of AI communication
Voice AI is transforming industries, redefining customer interactions, and creating new possibilities for automation. Whether you're building multilingual voice agents or embedding speech understanding into your product, Solaria-1 gives you a competitive edge with unmatched speed, accuracy, and language reach.
For developers, Solaria-1 offers plug-and-play APIs, real-time performance, and support for advanced customization—so you can build fast, scale effortlessly, and ship smarter.
For decision-makers, the model delivers enterprise-grade accuracy, global reach, and the infrastructure to support mission-critical applications—now and in the future.
Bonus feature: The story behind Solaria—or the sci-fi roots of our mission
Gladia is a character from Isaac Asimov’s novel The Naked Sun, a story set on the planet Solaria—a world defined by solitude, advanced technology, and the tension between human and robotic interaction. Gladia, the character, stands out as someone striving to bridge emotional and social gaps, breaking away from the norms of her society to foster deeper human connections.
These qualities that distinguish Gladia—empathy, inclusivity, and ethical consideration—are at the core of what we, as a company, aim to bring into our products and culture.
With the recent release of Solaria, a model designed to bridge linguistic divides across global markets, the company’s story has come full circle. The sci-fi inspiration that helped shape its visual identity continues to influence our mission: building tools that connect people, across languages, borders—and perhaps one day, even beyond.
The team at Gladia remains proud to carry this vision forward, and hope you join us on this cosmic journey to push the boundaries of innovation together.
Gladia Solaria, catching voice across the universe
Contact us
Your request has been registered
A problem occurred while submitting the form.
Read more
Speech-To-Text
Add speech-to-text to a Pipecat voice agent
Speech-To-Text
Add speech-to-text to a LiveKit voice agent
Speech-To-Text
Add speech-to-text to a Recall.ai meeting bot
From audio to knowledge
Subscribe to receive latest news, product updates and curated AI content.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
