# Gradium

> Gradium develops voice AI models for natural, expressive, low-latency voice interactions at scale.

Gradium provides Text-to-Speech, Speech-to-Text and speech-to-speech models built for real-time voice agents. This file indexes every published guide and blog post. The full text of each page is at https://gradium.ai/llms-full.txt.

Index generated 2026-10-21. Each entry carries the date its figures were last verified.

## Key resources

- [API documentation](https://docs.gradium.ai): endpoints, parameters and limits
- [Text to Speech](https://gradium.ai/text-to-speech): real-time text-to-speech with expressive voices
- [Speech to Text](https://gradium.ai/speech-to-text): streaming transcription with semantic turn detection
- [Voice Cloning](https://gradium.ai/voice-cloning): instant and pro voice clones
- [Live Translation](https://gradium.ai/live-translation): real-time speech translation between five languages, in the voice you choose
- [Pricing](https://gradium.ai/pricing): plans, credits and concurrency
- [Blog](https://gradium.ai/blog): research, engineering and product posts
- [Blog RSS feed](https://gradium.ai/blog/feed.xml)
- [Guides RSS feed](https://gradium.ai/content/feed.xml)
- [Studio](https://studio.gradium.ai/): voice playground and account management
- [Voices](https://gradium.ai/voices): the voice library, accents and the prompts that design a voice
- [Voice Designer](https://studio.gradium.ai/voices/design): generate an original voice from a description

## Guides, by topic

### Benchmarks

Independent and Gradium-run measurements of Text-to-Speech and Speech-to-Text latency and accuracy, each figure dated.

- [TTS Latency Benchmark 2026: Time to First Audio](https://gradium.ai/content/tts-latency-benchmark-2026) [hub]: TTS latency benchmark 2026: the Coval leaderboard on perceived time to first audio, with Gradium TTS at 214 ms median, September 2026. (last verified 2026-09-10)
- [Gradium vs ElevenLabs for Voice Agents: 2026 Benchmark](https://gradium.ai/content/gradium-vs-elevenlabs-voice-agents-benchmark): Gradium vs ElevenLabs, September 2026: Coval latency, three accuracy tests including an 81.0% hard-case pass rate, Elo and price per 1M characters. (last verified 2026-09-10)
- [STT API Benchmark 2026: Latency and Accuracy](https://gradium.ai/content/stt-api-benchmark-2026-latency-accuracy): Speech-to-Text for voice agents: time to first token against word error rate, why no public cross-vendor board exists, and how to run the test yourself. (last verified 2026-09-10)
- [TTS WER Benchmark 2026: Pronunciation Accuracy](https://gradium.ai/content/tts-wer-benchmark-2026): TTS WER benchmark 2026: Coval, a multilingual test set and an 81.0% hard-case pass rate, September 2026. Every figure dated and sourced. (last verified 2026-09-10)

### Choosing a Text-to-Speech API

Rankings, criteria frameworks and head-to-head comparisons for picking a Text-to-Speech provider.

- [Best Text-to-Speech APIs 2026: The Full Ranking](https://gradium.ai/content/best-text-to-speech-apis-2026) [hub]: Ten-plus Text-to-Speech providers ranked on dated September 2026 measurements: Coval latency, Artificial Analysis Elo, hard-case accuracy and price. (last verified 2026-09-10)
- [Best AI Voice Generators 2026 for Creators](https://gradium.ai/content/best-ai-voice-generators-2026): AI voice generators for creators, September 2026: blind preference Elo, an 81.0% accuracy test, cloning tiers and cost per hour of audio. (last verified 2026-09-10)
- [Best Speech APIs 2026: TTS and STT From One Vendor](https://gradium.ai/content/best-speech-apis-2026): Which vendors ship both Text-to-Speech and Speech-to-Text, what a one-vendor stack costs per hour and per 1M characters, September 2026. (last verified 2026-09-10)
- [Best Text-to-Speech API for Voice Agents 2026](https://gradium.ai/content/best-text-to-speech-api-voice-agents): The Text-to-Speech comparison for voice agents: turn budget, latency spread and hard-case pronunciation, with dated September 2026 figures. (last verified 2026-09-10)
- [Best TTS API 2026: The Short Verdict by Use Case](https://gradium.ai/content/best-tts-api-2026): Best TTS API 2026, refreshed monthly: one pick per use case with dated September 2026 figures from Coval, Artificial Analysis and a hard-case test. (last verified 2026-09-10)
- [How to Choose a TTS API: The Five Criteria](https://gradium.ai/content/how-to-choose-a-tts-api): How to choose a TTS API: five criteria to rank before you shortlist, with dated September 2026 figures from Coval and an 81.0% hard-case test. (last verified 2026-09-10)
- [ElevenLabs vs Cartesia vs OpenAI vs Gradium: TTS in 2026](https://gradium.ai/content/how-to-pick-tts-provider-elevenlabs-cartesia-openai): Four-way TTS head-to-head, September 2026: Coval latency, hard-case accuracy at 81.0%, Artificial Analysis Elo and price per 1M characters. (last verified 2026-09-10)
- [Top 3 Text-to-Speech Solutions 2026: Scored](https://gradium.ai/content/top-3-text-to-speech-solutions-2026): Gradium, ElevenLabs and Cartesia scored on seven dated criteria, September 2026, with two weightings for live agents and content production. (last verified 2026-09-10)

### Building Voice Agents

Stacks, orchestration frameworks and integration guides for production voice agents.

- [Best API to Build an AI Voice Agent in 2026](https://gradium.ai/content/best-api-build-ai-voice-agent-2026) [hub]: Best API to build an AI voice agent in 2026: orchestration frameworks and voice layers compared, with dated September 2026 Coval and accuracy figures. (last verified 2026-09-10)
- [Build an Audiobook Agent with Gradium and Pipecat](https://gradium.ai/content/audiobook-agent-gradium-pipecat): Build a real-time story narrator with Gradium Text-to-Speech and Pipecat: installation, pipeline, voice configuration and session limits, in about 100 lines. (last verified 2026-09-10)
- [Best Voice AI API for Phone-Based Voice Agents in 2026](https://gradium.ai/content/best-voice-ai-api-phone-based-voice-agents-2026): Voice AI APIs for phone agents: sample rates, codecs and the hard-case pronunciation that decides a call, with dated September 2026 figures. (last verified 2026-09-10)
- [Cascaded Voice Agents vs Speech-to-Speech in 2026](https://gradium.ai/content/cascaded-voice-agent-vs-speech-to-speech-2026): Cascade or speech-to-speech: latency, modularity, paralinguistic information and LLM flexibility compared, and why cascade still dominates production in 2026. (last verified 2026-09-10)
- [Gradium TTS in Pipecat: Setup and Integration Guide](https://gradium.ai/content/gradium-pipecat-native-tts-integration-voice-agents): Gradium TTS integration for Pipecat: install GradiumTTSService, configure voice and model settings, and stream low-latency speech in your voice agent. (last verified 2026-09-10)
- [How to Add Text-to-Speech to an App or Website](https://gradium.ai/content/how-to-add-text-to-speech-to-app-website-voice-agent): How to add TTS to your app, website, or voice agent: streaming WebSocket, LiveKit, Pipecat, and direct API integration with Gradium in under 100 lines. (last verified 2026-09-10)
- [Build a Voice AI Agent with Gradium and LiveKit](https://gradium.ai/content/how-to-build-voice-ai-agent-gradium-livekit): Build a conversational voice agent in about 100 lines of Python: the Gradium LiveKit plugin, semantic VAD, interruptions, preemptive generation and deployment. (last verified 2026-09-10)
- [How to Connect Text-to-Speech to an LLM Pipeline](https://gradium.ai/content/how-to-connect-tts-to-llm-voice-pipeline): How to connect TTS to an LLM in a voice pipeline: cascade architecture, streaming optimization, end-of-turn detection, and total latency budget explained. (last verified 2026-09-10)
- [How to Turn an LLM into a Voice Agent: Best Stack 2026](https://gradium.ai/content/turn-llm-into-voice-agent-best-stack-2026): Turn any LLM into a voice agent: the cascade stack, an orchestration framework and a voice layer, with dated September 2026 latency and accuracy figures. (last verified 2026-09-10)

### Latency and Streaming

Time to first audio, streaming protocols and connection reuse for real-time speech.

- [Best Low-Latency TTS APIs 2026](https://gradium.ai/content/best-low-latency-tts-apis-2026) [hub]: Compare the best low-latency TTS APIs in 2026, benchmarked by TTFA (P50/P99), WebSocket architecture, codebook configs and full voice agent pipeline impact. (last verified 2026-09-10)
- [How to Stream TTS Audio in Real Time](https://gradium.ai/content/how-to-stream-tts-audio-real-time-bidirectional-streaming): How to stream TTS in real time: WebSocket lifecycle, bidirectional streaming, LLM-TTS interleaving, and multiplexing explained with Gradium's API. (last verified 2026-09-10)
- [WebSocket Multiplexing for Gradium TTS](https://gradium.ai/content/multiplexing-tts-websocket-gradium): Reuse one WebSocket across many synthesis requests: close_ws_on_eos, request routing, a measured 44 ms saving at the median, and the four ways it breaks. (last verified 2026-09-10)

### Pronunciation and Accuracy

Getting names, acronyms, numbers, dates and currencies spoken correctly.

- [How to Make TTS Pronounce Numbers and Dates Correctly](https://gradium.ai/content/how-to-make-tts-pronounce-numbers-dates-currencies-phone-numbers) [hub]: Get numbers, dates, currencies and phone numbers spoken correctly: rewrite_rules, custom pronunciations, and an 81.0% hard-case test from August 2026. (last verified 2026-09-10)
- [How to Fix TTS Mispronouncing Names and Acronyms](https://gradium.ai/content/how-to-fix-tts-mispronouncing-names-acronyms-technical-terms): How to fix TTS mispronouncing names, acronyms, and technical terms: custom pronunciations in Gradium Studio and the Python SDK, step by step. (last verified 2026-09-10)
- [How to Stop TTS Switching Accents Mid-Sentence](https://gradium.ai/content/how-to-stop-tts-switching-accents-languages-mid-sentence): Why TTS switches accents or languages mid-sentence, and how to fix it: native code-switching, single-model architecture, and language depth explained. (last verified 2026-09-10)
- [json_config in Gradium: TTS and STT Parameters](https://gradium.ai/content/how-to-use-json-config-gradium-tts-stt): Every json_config parameter for Gradium Text-to-Speech and Speech-to-Text, with ranges, defaults, code samples, and the two fields that are not in it. (last verified 2026-09-10)
- [Custom Pronunciation in Gradium TTS](https://gradium.ai/content/pronunciation-dictionaries-gradium-tts): Control how names, acronyms and brand terms are spoken without altering your input text: creating a custom pronunciation, pronunciation_id, and testing it. (last verified 2026-09-10)
- [Text Normalization for TTS Edge Cases](https://gradium.ai/content/text-normalization-tts-edge-cases-gradium): Dates, numbers, emails, URLs and codes spoken correctly: rewrite_rules, language aliases against explicit rule lists, and how to order them. (last verified 2026-09-10)

### Voice Cloning

Instant and professional cloning, accents, and keeping one voice consistent across sessions.

- [Best Voice Cloning APIs 2026: Instant vs Professional](https://gradium.ai/content/best-voice-cloning-apis-2026) [hub]: Compare the best voice cloning APIs in 2026: Gradium, ElevenLabs, Cartesia. Instant vs. professional cloning, speaker similarity benchmarks, pricing. (last verified 2026-09-10)
- [How to Clone a Voice or Create a Custom Voice with Gradium](https://gradium.ai/content/how-to-clone-a-voice-create-custom-voice): How to clone a voice with Gradium: Instant Voice Clone from 10 seconds, Pro Voice Clone from 30 minutes, the 3,220-pair benchmark, consent and data handling. (last verified 2026-09-14)
- [How to Make AI Voices Sound More Natural and Less Robotic](https://gradium.ai/content/how-to-make-ai-voices-sound-more-natural): How to make AI voices sound more natural: micro-traits, cfg_coef and padding_bonus in json_config, register matching, and a 3,220-pair blinded benchmark. (last verified 2026-09-14)
- [Keeping a Voice Consistent Across Sessions](https://gradium.ai/content/how-to-keep-voice-consistent-long-form-multi-session-audio): How to keep a voice consistent across long-form or multi-session audio: voice cloning tiers, micro-traits, session architecture, and cross-session identity. (last verified 2026-09-10)
- [Instant vs Pro Voice Cloning in Gradium: When to Use Each](https://gradium.ai/content/instant-vs-pro-voice-cloning-gradium): Instant or Pro voice cloning in Gradium: 10 seconds against 30 minutes, what each trades away, cost per character, and how to choose. September 2026. (last verified 2026-09-10)
- [Why Generic TTS Fails on Regional Accents](https://gradium.ai/content/voice-cloning-regional-accents-2026): Regional accent voice AI: generic TTS flattens Bavarian, Argentinian, and Quebecois speech into one default. Gradium preserves accent from a 10-second sample. (last verified 2026-09-10)

### On-Device Text-to-Speech

Gradium Phonon and the tradeoffs between cloud and on-device synthesis.

- [On-Device Text-to-Speech: How Edge Synthesis Works](https://gradium.ai/content/on-device-text-to-speech-2026) [hub]: On-device Text-to-Speech explained: how edge synthesis works, what a 100M-parameter model trades away, and when it beats a cloud API. September 2026. (last verified 2026-09-10)
- [Cloud vs On-Device TTS: The Decision Guide](https://gradium.ai/content/cloud-vs-on-device-tts-how-to-choose): Cloud or on-device Text-to-Speech: the four conditions that push a product to the edge, and what you give up. Updated September 2026. (last verified 2026-09-10)
- [Gradium Phonon: On-Device Text-to-Speech](https://gradium.ai/content/gradium-phonon-on-device-tts): Gradium Phonon, the on-device Text-to-Speech model: roughly 100M parameters, five languages since July 2026, licensing and integration. (last verified 2026-09-10)
- [Gradium Phonon Benchmarks: Results by Release](https://gradium.ai/content/gradium-phonon-on-device-tts-benchmarks-2026): Phonon's own results release by release: 1.48% word error rate in April 2026, 1.00% in May, and 1.07% multilingual in July 2026. (last verified 2026-09-10)
- [On-Device TTS Benchmark 2026: Phonon vs Kani vs NeuTTS](https://gradium.ai/content/on-device-tts-benchmark-2026): On-device TTS benchmark: Phonon at 1.48% word error rate against NeuTTS Nano 1.71%, NeuTTS Air 2.18% and Kani TTS 2 4.97% on Seed-TTS, April 2026. (last verified 2026-09-10)
- [Phonon on Seed-TTS: The May 2026 Record](https://gradium.ai/content/phonon-seed-tts-benchmark-2026): A dated record of one run: Gradium Phonon at 1.00% word error rate on the Seed-TTS English evaluation, measured May 26, 2026. (last verified 2026-09-10)

### Provider Alternatives

Side-by-side migration guides from ElevenLabs, Cartesia, Deepgram and Azure.

- [Best ElevenLabs Alternatives 2026: Quality vs Price](https://gradium.ai/content/best-elevenlabs-alternatives-2026-tts-apis-voice-quality-price) [hub]: ElevenLabs alternatives ranked on blind-preference Elo against price, September 2026, with the hard-case accuracy test where the order changes. (last verified 2026-09-10)
- [Azure TTS Alternative: Gradium for Real-Time Voice AI](https://gradium.ai/content/azure-tts-alternative-gradium): Azure AI Speech alternative: Gradium on hard-case pronunciation (81.0%, August 2026), latency, deployment range and price. Dated September 2026 figures. (last verified 2026-09-10)
- [Cartesia Alternative: Gradium for Real-Time Voice AI](https://gradium.ai/content/cartesia-alternative-gradium-voice-ai): Cartesia vs Gradium, September 2026: preference Elo, Coval latency, an 81.0% hard-case pronunciation test, cloning and deployment range compared. (last verified 2026-09-10)
- [Deepgram Alternative: Gradium for Real-Time Voice AI](https://gradium.ai/content/deepgram-alternative-gradium-voice-ai): Deepgram vs Gradium, September 2026: Speech-to-Text price per hour, Coval latency and spread, voice cloning, and one credit pool across both halves. (last verified 2026-09-10)
- [ElevenLabs Alternative: Gradium for Voice Agents](https://gradium.ai/content/elevenlabs-alternative-gradium-voice-ai): ElevenLabs vs Gradium, September 2026: median latency, an 81.0% hard-case pronunciation test, price per 1M characters and Speech-to-Text billing. (last verified 2026-09-10)

### Turn-Taking and VAD

Voice activity detection, semantic turn detection and interruption handling.

- [Turn-Taking in Voice Agents: Why VAD Is Hard](https://gradium.ai/content/turn-taking-voice-agents-vad) [hub]: Why turn-taking breaks voice agents: energy thresholds, silence timers and what each gets wrong. Human turns change hands in about 208 ms. (last verified 2026-09-10)
- [Semantic VAD: Turn Detection in Gradium STT](https://gradium.ai/content/semantic-vad-voice-agents-turn-detection-2026): Gradium's Semantic VAD: end-of-turn detection from meaning rather than silence duration, native to its Speech-to-Text, and how to configure it. (last verified 2026-09-10)

### Multilingual Speech

Cross-language synthesis, code-switching and per-language accuracy.

- [Best Multilingual TTS APIs 2026: Breadth vs Depth](https://gradium.ai/content/best-multilingual-tts-apis-2026) [hub]: Compare the best multilingual TTS APIs in 2026: Gradium, ElevenLabs, Cartesia, Deepgram on language coverage, code-switching, voice cloning and latency. (last verified 2026-09-10)

### Pricing and Scale

Per-character and per-hour cost comparisons, credits, and concurrency limits.

- [How to Compare TTS Pricing Across Providers in 2026](https://gradium.ai/content/how-to-compare-tts-pricing-across-providers-2026) [hub]: TTS pricing in 2026: per-character, credit-based, token-based models explained. Full comparison of Gradium, ElevenLabs, Cartesia, OpenAI, and more. (last verified 2026-09-10)
- [How to Estimate TTS Costs at Scale: A Practical Guide](https://gradium.ai/content/how-to-estimate-tts-costs-at-scale-credits-concurrency): How to estimate TTS costs at scale: credit conversion, concurrency limits, a worked example at $508 a month, and a September 2026 provider comparison. (last verified 2026-09-14)

## Blog

- [Gradium releases Phonon, an open-source text-to-speech model that runs on any device](https://gradium.ai/blog/gradium-releases-phonon): Trained on NVIDIA GPUs, Phonon generates natural speech locally in as little as 64.5 MB, on phones, laptops, embedded hardware and in the browser. (last verified 2026-10-21)
- [Open-Sourcing Phonon: A Suite of Frontier On-Device TTS Models](https://gradium.ai/blog/release-phonon-open-weight): Today, we're open-sourcing Phonon, now a family of 31m and 40m parameter text-to-speech (TTS) models that set a new standard for naturalness and accuracy on-device. (last verified 2026-10-20)
- [Gradium is now available on Eden AI](https://gradium.ai/blog/eden-ai-gradium-partnership): Eden AI is an AI gateway: one API for 500+ models across LLMs, speech, vision and translation, with routing, fallbacks and a dedicated EU endpoint for European data residency. Gradium's Text-to-Speech, Speech-to-Text and speech translation are now available through that API, pinned to the EU region. (last verified 2026-10-19)
- [1,000+ new voices, our most natural set yet](https://gradium.ai/blog/more-than-1000-voices): The Gradium voice catalog grows from ~380 to 1.4k voices. The 1,000+ new voices cover 28 accents across English, French, Spanish, German and Portuguese, cast for customer service, narration, ads and characters. Listen to 62 of them, and see how they were chosen. (last verified 2026-10-07)
- [Why we built the fastest robust TTS model](https://gradium.ai/blog/fastest-robust-tts-model): Our latest Text-to-Speech model takes under 50ms for the first audio chunk to arrive. Choosing a fast model does not have to come with cutting corners for model robustness: it improves naturalness and expressivity while retaining our strengths on hard cases like phone numbers and alphanumeric text. (last verified 2026-10-01)
- [Giving voice agents a voice that adapts to context](https://gradium.ai/blog/voice-that-adapts-to-context): Most voice agents speak with one preselected voice, whoever is on the line. Once a voice can be created from a text description, the voice becomes another property your code computes from context. This article builds the full pipeline: classify the message, map the signals to a Voice Design prompt, resolve it to an approved voice on the hot path, and design the missing ones in the background. (last verified 2026-09-29)
- [How a benchmark change produced a faster TTS model](https://gradium.ai/blog/coval-perceived-ttfa-benchmark): On June 3, 2026, Gradium's time to first audio on the Coval TTS leaderboard moved from 171.9 ms to 429.6 ms. No model had shipped and no serving infrastructure had changed. What changed was the measurement: Coval had started counting the silence at the start of the stream. (last verified 2026-09-09)
- [Gradium now available for free on LiveKit Inference](https://gradium.ai/blog/livekit-gradium-partnership): Gradium Text-to-Speech is now available on LiveKit Inference, free until Oct. 9. Gradium TTS is designed for conversational use cases, with voice cloning, low latency (below 250ms TTFA), and robust pronunciation on difficult cases. (last verified 2026-09-09)
- [Launching Voice Design: prompt the voice your agent needs](https://gradium.ai/blog/launching-voice-design): Developers building voice agents ask us for voices matched to the use case in front of them: a Québécoise receptionist for a Montréal dealership, a Paulista support agent for a São Paulo fintech, a narrator in his sixties with the authority of a lecture hall. Briefs outnumber any catalog. Voice Design starts from a prompt: write one or two sentences, get complete new voices back in seconds, keep the one you want. (last verified 2026-09-07)
- [Build voice agents with EU and US residency and Zero Data Retention](https://gradium.ai/blog/eu-data-residency-zero-data-retention): Two questions come up in almost every enterprise evaluation: where exactly does the inference run, and what do you keep? Data residency pins your Text-to-Speech, Speech-to-Text and Speech-to-Speech sessions to our EU servers, or to our US servers, enforced at the server and reported on every response, so you can assert on it instead of trusting a hostname. Zero Data Retention decides whether the text and audio processed by our servers survive the call at all. (last verified 2026-09-02)
- [Gradium TTS: you no longer have to choose between latency and accuracy](https://gradium.ai/blog/gradium-tts-latency-and-accuracy): A new Gradium TTS model is available today, and it is now the default. It reads the cases that break voice agents in production, phone numbers, email addresses, IBANs, with no pre-processing or text normalization on your side. Time to first audio (TTFA) P50 is 216 ms on Coval, 170 ms faster than the model it replaces, with a 30 ms p75 - p25 spread, the tightest of the five models tested. Here is the hard-case evaluation set, the head-to-head samples, and how to hear the difference yourself. (last verified 2026-08-31)
- [Gradium voices are now live on AudioStack](https://gradium.ai/blog/audiostack-gradium-partnership): AudioStack, the agentic audio production platform for media and advertising, has added Gradium to its roster of voice providers. Our voices are live on the platform now, across multiple languages and regional accents. (last verified 2026-08-19)
- [New voices: how we pick the one that wins](https://gradium.ai/blog/new-voices-how-we-pick-the-one-that-wins): 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. (last verified 2026-08-04)
- [Gradium TTS public beta: open for your hardest cases](https://gradium.ai/blog/gradium-tts-public-beta): 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. (last verified 2026-07-30)
- [Keyword boosting: teaching Speech-To-Text your vocabulary](https://gradium.ai/blog/stt-keyword-boosting): 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. (last verified 2026-07-21)
- [Smaller, clearer, better: Phonon Multilingual beats Magpie and NeuTTS Nano in French, German, and Spanish](https://gradium.ai/blog/phonon-multilingual-smaller-clearer-better): 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. (last verified 2026-07-15)
- [Natural conversation, live retrieval: Keenable search with Gradium](https://gradium.ai/blog/keenable-gradium-partnership): 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. (last verified 2026-07-10)
- [Gradium Extends Funding to $100 Million and Expands to Silicon Valley](https://gradium.ai/blog/gradium-100-million-funding-nvidia): 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. (last verified 2026-07-08)
- [Gradium Powers RMC BFM Drive: AI-Generated Personalized Radio in Renault Vehicles](https://gradium.ai/blog/rmc-bfm-drive-gradium): 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. (last verified 2026-07-07)
- [Launching Gradium Translate: the best accuracy-latency tradeoff against gemini-3.5-live-translate and gpt-realtime-translate](https://gradium.ai/blog/launching-gradium-translate): 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. (last verified 2026-06-24)
- [Gradium TTS, upgraded: more accurate Text-To-Speech](https://gradium.ai/blog/gradium-tts-upgrade): 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. (last verified 2026-06-10)
- [Semantic VAD: turn detection that uses meaning, not silence](https://gradium.ai/blog/semantic-vad): 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. (last verified 2026-06-02)
- [Phonon update: 1.00% WER on Seed-TTS, smaller than every model we beat](https://gradium.ai/blog/phonon-update-may-2026): 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. (last verified 2026-05-26)
- [Gradium #1 on Coval TTS Benchmarks](https://gradium.ai/blog/coval-tts-benchmarks-may-2026): 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. (last verified 2026-05-13)
- [Gradium Voice Launches on AWS as a SaaS Subscription and a SageMaker Model Image](https://gradium.ai/blog/gradium-aws-launch): 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. (last verified 2026-04-29)
- [The most accurate multilingual text-to-speech, by the numbers](https://gradium.ai/blog/word-error-rate-evaluations): 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. (last verified 2026-04-29)
- [Gradbot: Vibe code voice agents in 50 lines of code](https://gradium.ai/blog/gradbot): 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. (last verified 2026-04-15)
- [Evaluating Phonon: how we made the best TTS model for edge devices](https://gradium.ai/blog/evaluating-phonon): An evaluation of Gradium Phonon, our on-device text-to-speech model. Despite its small size, it significantly outperforms larger models. (last verified 2026-04-09)
- [Gradium Phonon: On-Device TTS for Consumer Apps, NPCs, and Offline Products](https://gradium.ai/blog/gradium-phonon): Announcing Gradium Phonon, our new on-device text-to-speech model designed for consumer apps, NPCs, and offline products. (last verified 2026-04-03)
- [Time to First Audio: Measuring and Reducing TTS Latency in Voice Agents](https://gradium.ai/blog/time-to-first-audio): 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. (last verified 2026-03-24)
- [InteractionLabs (Ongo) and Gradium Partner to Redefine Human-Robot Interaction](https://gradium.ai/blog/interactionlabs-gradium-partnership): InteractionLabs, the company behind the Ongo living lamp robot, and Gradium announce a partnership to bring expressive, real-time voice AI to robotics. (last verified 2026-02-17)
- [Optimizing Quality vs. Latency in Real-Time Text-to-Speech AI Models](https://gradium.ai/blog/optimizing-quality-vs-latency): 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. (last verified 2026-02-11)
- [Building Voice Agents From the Ground Up: The Gradium Startup Program](https://gradium.ai/blog/gradium-startup-program): 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. (last verified 2026-02-09)
- [Acolad and Gradium Partner to Advance Enterprise-Ready AI Interpreting](https://gradium.ai/blog/acolad-gradium-partnership): 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. (last verified 2026-02-05)
- [Invincible Voice: How Gradium's Real-Time Voice AI Helps ALS Patients Speak Again](https://gradium.ai/blog/invincible-voice): Gradium's voice AI technology powers Invincible Voice, an open-source assistive system helping people with ALS and speech loss communicate in real-time. (last verified 2026-01-28)
- [Why Your Voice Cloning Sounds Fake (And How to Fix It)](https://gradium.ai/blog/voice-cloning-sounds-fake): Discover how Gradium's instant voice cloning achieves superior speaker similarity to ElevenLabs. Benchmark results across 4 languages with 3,220 human evaluations. (last verified 2026-01-22)
- [Powering Wonderful's Voice Agents](https://gradium.ai/blog/wonderful): We're proud to power real-time voice agents on Wonderful's platform, bringing cutting-edge voice AI from experimental to deployable. (last verified 2026-01-15)
- [Gradium: Solving voice](https://gradium.ai/blog/gradium): Today we're excited to launch Gradium, the core engine powering the next generation of voice products and interactions. (last verified 2025-12-02)

## About Gradium

- Product: voice AI platform (Text-to-Speech, Speech-to-Text, speech-to-speech)
- Architecture: Delayed Streams Modeling ("Streaming Sequence-to-Sequence Learning with Delayed Streams Modeling", arXiv:2509.08753)
- Current cloud model: Gradium TTS, released August 31, 2026, API `model_name: "default"`, LiveKit Inference id `gradium/default`
- Independent measurement: on the Coval TTS leaderboard (September 10, 2026, 1-day window, 27 models) Gradium TTS records a 214 ms perceived time to first audio and 0 ms leading silence, the only model on the board at zero. Live board: https://benchmarks.coval.ai/tts
- Pronunciation: 81.0% pass rate on a 500-sentence hard-case set across five languages (Gradium, August 2026); dataset at https://huggingface.co/datasets/gradium/tts-eval-customer-support-202608
- Multilingual accuracy: 1.11% average word error rate on the MiniMax Multilingual TTS Test Set across English, French, Spanish, Portuguese and German (Gradium, April 2026)
- Languages: English, French, German, Spanish and Portuguese, with mid-sentence code-switching
- Speech-to-Text: streaming, with native semantic voice activity detection, on the same credit pool as Text-to-Speech
- On-device: Gradium Phonon, a roughly 100M-parameter on-device Text-to-Speech model
- Deployment: cloud API, dedicated instances, self-hosted, on-premise, on-device
- Session limit: 3,000 seconds per session, Text-to-Speech and Speech-to-Text
- Free tier: 45,000 credits per month, no credit card required