Launching today
Soniox STT v6
Speech recognition built for the next generation of voice AI
1 follower
Speech recognition built for the next generation of voice AI
1 follower
Soniox STT v6 is a speech recognition model for voice agents, meetings, and voice applications. It understands agent-side context, filters out the agent's own voice, and delivers more accurate real-time transcription across 60+ languages.
Payment Required
Launch Team
Atlassian LoomShow, don't just tell with Loom
Promoted
Voice agents need to understand not just what users say, but also the context of the conversation.
Soniox STT v6 makes this possible with improved speech recognition and two new capabilities designed specifically for voice agents.
Key features:
Agent text input: Give the speech model context from both sides of a conversation to recognize short answers, names, numbers, and specialized terms more accurately.
Agent voice echo cancellation: Automatically filters out the AI agent's own spoken voice while continuing to listen to the user, even when they interrupt.
Improved transcription: Better recognition of numbers, currencies, dates, names, addresses, and specialized terminology.
Speaker diarization: More accurately identifies who said what, even with overlapping speech and interruptions.
60+ languages: Supports multilingual conversations, accents, and mid-sentence language switching.
Real-time translation: More accurate speech translation while preserving important details like amounts, dates, and identifiers.
Real-time and async APIs: Process live conversations or transcribe recorded audio.
Who it's for: Developers building voice agents, AI meeting assistants, dictation tools, transcription products, and multilingual voice applications.
Explore Soniox STT v6 | API docs