Picking a voice works like casting: one candidate in a thousand has a hit factor you cannot explain. Here is the methodology behind every flagship voice in the Gradium catalog, from generating hundreds of candidates per category to ranking them head-to-head against the incumbents on an ELO benchmark, why the same prompt does not travel across locales, and how to write the target yourself.
A new Gradium TTS model is available today in public beta. It keeps improving on the current production model, handling the complex cases natively, phone numbers, email addresses, IBAN numbers, time expressions and more, with no pre-processing or text normalization on your side. Anyone can test it through the API, and everyone who sends us feedback during the beta receives 1M Gradium credits.
Speech-To-Text models rarely see your product's vocabulary, brand names, drugs, athletes, and places, so they fall back on a similar-sounding common word. Keyword boosting raises the probability of the terms you care about while decoding, in real time, with no retraining. Here is how the boost parameter behaves, what a sweep across hard clips shows, and how to add it in a few lines.
Phonon, our 100M-parameter on-device Text-To-Speech model, produces up to 3.5x fewer word errors than NVIDIA's Magpie TTS in French, German, and Spanish at 3.6x fewer parameters, and leads NeuTTS Nano on both word error rate and speaker similarity. It clones a reference voice across five languages: English, French, German, Spanish, and Portuguese.
Gradium and Keenable are partnering to bring realtime web search into natural voice agents. Gradbot, Gradium's open source voice agent framework, is adding Keenable search so agents can retrieve fresh answers from the live web without breaking the flow of conversation.
Gradium extends its seed funding to $100 million, welcoming new investors including NVIDIA, and opens a San Francisco Bay Area office to scale its real-time voice AI.
Gradium powers RMC BFM Drive, the first fully AI-generated, personalized radio experience built into Renault connected vehicles. Our Text-To-Speech models synthesize transitions in real time using Pro Voice Clones of RMC BFM journalists, turning an AI-curated program into something that sounds and feels like a live broadcast.
Gradium launches stt-translate and s2s-translate: real-time Speech-To-Text and Speech-To-Speech translation across English, French, German, Spanish, and Portuguese, with a better accuracy-latency tradeoff than gpt-realtime-translate and gemini-3.5-live-translate and full control of the output voice, including cloning.
Gradium TTS now runs on a new model: more natural prosody and substantially more accurate pronunciation on the cases that break voice agents in production, including spelling, acronyms, emails, phone numbers, and codes. It wins head-to-head against our previous model in all five languages, and leads real-time competitors (Cartesia Sonic 3.5, Inworld TTS 1.5 Max, ElevenLabs Flash v2.5 and Multilingual v2) on the hardest pronunciation cases. Available now as the new default, with custom voices carried over.
Acoustic VAD answers "is there a voice right now?" Semantic VAD answers "is the user done talking?" Here's why the distinction decides whether a voice agent cuts users off, how Gradium STT emits multi-horizon turn-completion predictions every 80 ms, and how to tune delay_in_frames, horizon, and flushing for your use case.
Phonon, our 100M-parameter on-device Text-To-Speech model, reaches 1.00% WER on the Seed-TTS English benchmark, outperforming NeuTTS Air, KaniTTS2, and NeuTTS Nano. With a fixed voice, it drops to 0.83% WER, ahead of Kokoro and Magpie.
Independent Coval TTS benchmark (May 2026): Gradium ranks first on P50 TTFA (158ms), latency IQR (2ms), and provides SOTA WER (3.7%) against ElevenLabs Turbo v2.5, Flash v2.5, Multilingual v2, Cartesia Sonic-3, Deepgram Aura-2, Rime Mist-v3, Arcana, and OpenAI TTS-1-HD.
Gradium is now available on AWS through two paths: a fully managed SaaS subscription via AWS Marketplace, and a deployable model image via Amazon SageMaker for teams that need in-VPC inference.
How we measure WER for TTS at Gradium: text normalization, jiwer alignment, results on the MiniMax Multilingual benchmark across English, French, Spanish, Portuguese and German — and why the standard metric is starting to saturate.
Gradbot is our open-source framework for prototyping voice agents in minutes. Built on a Rust orchestration core, it handles turn-taking, interruptions, silence, and async tool calls so you can ship a working voice experience in around 50 lines of code.
An evaluation of Gradium Phonon, our on-device text-to-speech model. Despite its small size, it significantly outperforms larger models.
Announcing Gradium Phonon, our new on-device text-to-speech model designed for consumer apps, NPCs, and offline products.
In natural conversation, the gap between one person finishing a sentence and the other starting to respond averages around 200 milliseconds. For voice agents this is the target to match.
InteractionLabs, the company behind the Ongo living lamp robot, and Gradium announce a partnership to bring expressive, real-time voice AI to robotics.
Explore strategies for balancing quality and latency in real-time TTS AI models. Learn how Gradium achieves low-latency, high-quality speech synthesis for voice applications.
Get 6 months free access to Gradium's voice AI platform. 9M monthly credits, voice cloning, STT/TTS APIs for seed-funded startups building voice-first products.
Acolad, the global leader in language and content solutions, and Gradium just announced a strategic partnership. The partnership reflects Acolad’s commitment to delivering secure, scalable, and governed AI-powered interpreting solutions, designed for enterprise and public-sector environments.
Gradium's voice AI technology powers Invincible Voice, an open-source assistive system helping people with ALS and speech loss communicate in real-time.
Discover how Gradium's instant voice cloning achieves superior speaker similarity to ElevenLabs. Benchmark results across 4 languages with 3,220 human evaluations.
We're proud to power real-time voice agents on Wonderful's platform, bringing cutting-edge voice AI from experimental to deployable.
Today we're excited to launch Gradium, the core engine powering the next generation of voice products and interactions.