Soniox@soniox_aiMay 5, 2026
Soniox published a nice blog post today about implementing multi-lingual voice agents.
For the last two years, we've heard over and over from our customers that it is hard to achieve really good transcription accuracy and natural speech for non-English voice agents.
Fortunately, that's changing! Soniox models support 60 languages, has excellent latency, and is widely deployed for voice agent use cases.
One point the blog post makes is that it's helpful to use transcription and voice models from the same provider. I think that's true, for several reasons.
One non-obvious reason is that building and evaluating in the context of a complete voice agent pipeline is important. Model providers that are trying to solve both transcription and voice generation challenges have a much better complete picture of the "full stack" engineering challenges we face, building production voice agents and deploying them at scale.
The blog post includes sample code for using Soniox models in the open source @pipecat_ai voice agent framework.
Soniox@soniox_aiSoniox Text-to-Speech is now fully released in @pipecat_ai.
With Soniox STT + TTS, developers can build real-time voice bots that understand and speak in 60+ languages.
Native-speaker accuracy. Accurate alphanumerics. Production scale.
Build global voice bots with Pipecat +