Soniox

Soniox v5 Real-Time is now available. Live speech AI is not batch transcription with lower latency. It has to turn raw, noisy, continuous audio into structured intelligence while people speak. What’s new: • Higher accuracy across 60+ languages • Completely reengineered speaker separation • Better spoken language identification • Higher-quality real-time translation across 3,600+ language pairs • Faster semantic endpointing for voice agents • Better alphanumeric recognition • More robust native context handling Built for voice agents, meetings, captions, translation, dictation, customer support, contact centers, and multilingual products. Read more: soniox.com/blog/soniox-v5…

Introducing Soniox Compare. Test speech-to-text, text-to-speech, and speech translation across leading providers using your own data. Compare real APIs side by side, review public pricing, and inspect the open-source code. soniox.com/blog/introduci…

Soniox is the complete voice AI platform. Speech-to-text. Text-to-speech. Speech translation. All built to the same standard. All available across the same 60+ languages. All through one API. No stitching together multiple vendors. No inconsistent language coverage. No tradeoff between quality, latency, and scale. World-leading speech-to-text. Natural real-time text-to-speech. Fast, multilingual speech translation. Built for production from the first API call to global enterprise scale. One platform for every voice AI application. soniox.com

We have invested enormous engineering effort into making Soniox more than world-leading voice AI. It is global infrastructure companies can trust for their most critical applications. Build locally. Deploy globally. Serve the world. soniox.com/docs/data-resi…

Behind every API request is a global production system engineered for high availability, automated scaling, fault isolation, observability, and reliable real-time processing at scale.

World-leading AI is not enough. To power production applications globally, the infrastructure behind the AI must be just as formidable. 🧵 Soniox is fully deployed across three major global regions: 🇺🇸 United States 🇪🇺 European Union 🇯🇵 Japan

Drug names, SKUs, invented words. Skip common vocabulary, the model has that covered. All of it fits in about 10,000 characters. Enough to onboard the model into your world. Learn more: soniox.com/docs/stt/conce…

Restaurant taking phone orders? Set "restaurant: Spice India, location: London, setting: phone ordering" in general, add your menu in terms. The model already knows what paneer tikka is. What it can't know is the signature dish name you invented last month, and now it will hear this dish exactly like in the menu. Put only the hard stuff in terms.

Soniox STT is smart out of the box. Medical terms, legal jargon, technical vocabulary, it knows all of that. But when a pilot calls out "TOGA," a model without context hears a toga. Your internal product names don't stand a chance either. 🧵

Streaming TTS becomes much more powerful when you know exactly when each part of the text was spoken. Soniox Text-to-Speech now supports character-level timestamps alongside streaming audio. For each character in the generated text, Soniox returns the start and end time of the audio that pronounces it. This lets you align playback with the exact text that produced it as the stream arrives. That means you can build live subtitles, caption highlighting, karaoke-style reading experiences, and better voice agent interruption handling. API features like this are what make voice AI products feel responsive, polished, and real. Watch the text highlight in sync with the generated voice.

@lkcv_7299 🐐🐐🐐

@soniox_ai are the goats for Spanish STT 😍 looking good, first impressions

Soniox Endpoint Detection delivers.

@AlecInRealTime Lowest latency is part of the product. Applies to shipping too. 🚀

@soniox_ai Y'all are some of the fastest moving people in the voice space right now ‼️

Soniox Text-to-Speech now supports: • Speed control • Character-level timestamps Speed control is learned at the model level, not applied as audio post-processing, so speech stays natural. Timestamps let you sync generated audio with text in real time, useful for live captions and voice agent interruption handling. Try it in the Soniox Playground: console.soniox.com/playground/tex… Related docs: soniox.com/docs/tts/conce… soniox.com/docs/tts/rt/ti…

Upload the sample through Soniox API or in the Soniox Console, get the voice ID, and use it like any other voice in the API. Docs: soniox.com/docs/tts/conce…

Read something plain and declarative. No questions, no exclamations. Hesitations and restarts get cloned too. About 60 words fills 20 seconds at a normal pace.

Advice on voice cloning Soniox TTS can clone a voice from a short (up to 20 seconds) sample. The clone works reliably in all 60+ supported languages, but only if the recording is good. How to make a good recording: 🧵

Speak in a steady tone for up to 20 seconds. Large swings in volume or pitch make the clone unstable. The tone you record is the tone the clone speaks in, in every language.