Dispatch
NVIDIA open-weights Magpie TTS Multilingual — 12 languages, 364M params, 32ms time-to-first-audio on B200
Magpie TTS Multilingual covers English, Spanish, French, German, Italian, Vietnamese, Mandarin, Hindi, Japanese, Modern Standard Arabic, Korean and Brazilian Portuguese with male and female voices per language, at 364M parameters. Latency: 32ms time-to-first-audio single-stream on B200 (239ms at 64 concurrent streams), 47ms/275ms on H100, 79ms/395ms on A100. Character error rates improved this release — French 2.70% → 1.54%, Spanish 1.14% → 0.60%. Weights ship on Hugging Face under the NVIDIA Open Model License with NIM containers for production, so voice-agent builders can run the whole thing on owned infrastructure instead of paying managed-service latency.
↳ Follow the thread