
Summarize with AI
What Gradium is
Gradium provides streaming models for text-to-speech, speech-to-text, and live speech-to-speech translation. Developers can use voices from its catalog, design a new voice from a text prompt, or clone an authorized recording through Gradium Studio and the API.
Funding snapshot
Gradium announced in July 2026 that it had extended its seed financing to $100 million, with NVIDIA among the new investors.
Key features
- Streaming text-to-speech with word-level timestamps. Gradium reports about 50 ms to first audible output and 0 ms of leading silence in its cited Coval benchmark.
- Streaming speech-to-text with keyword boosting and semantic voice activity detection to help agents tell when a caller has finished speaking.
- Speech-to-speech translation across English, French, German, Spanish, and Portuguese.
- About 1,400 catalog voices across 28 accents, including regional variants such as Québécois French and Brazilian Portuguese.
- Pronunciation controls for phone numbers, emails, dates, currencies, URLs, and confirmation codes, with rewrite rules and custom dictionaries for names and specialist terms.
- Prompt-based Voice Design, Instant Voice Cloning from a short sample, and Pro Voice Cloning for longer recordings.
Where it lives
Alternatives to consider
Where Gradium fits
Gradium supplies the speech layer for applications such as support lines, booking assistants, and live interpreters. A phone agent can use Gradium to transcribe a caller, pass the transcript to its language model, then stream the reply back to the caller. The agent still needs its own business logic and, for phone calls, a telephony connection. Gradium Studio provides a browser playground for trying voices before integrating the API.
How the free tier works
The free tier includes 45,000 monthly credits, about one hour of text-to-speech or a little over four hours of transcription if you use the full allowance for one service. Text-to-speech, transcription, and translation share the same balance and use credits at different rates. Paid plans start at $13 per month and add commercial use and higher concurrency.
Deployment and data controls
Paid plans offer EU or US processing and optional zero data retention for speech request and response payloads. Gradium lists ISO 27001:2022 certification and a SOC 2 Type II report in its Trust Center. Enterprise customers can request dedicated, self-hosted, or on-premises deployments. For offline synthesis, Gradium also offers the separate Phonon models, including an English model under Apache 2.0 and a multilingual model under a non-commercial license.
Sources & media
Weekly digest
Don't pick tools from a launch tweet.
Directory updates and practical AI picks — short, useful, unsubscribe anytime.