Foundation models
We build foundation models for voice, including models that understand and produce codeswitched speech.
We’re a research-forward organization building custom voice models for regional languages and mixed speech.Our goal is to make the voice interface of technology accessible to everyone.
We started our journey with Zero STT, the world’s most accurate transcription model, and expanded intocomplete voice agents that can be deployed in air-gapped environments.
We build foundation models for voice, including models that understand and produce codeswitched speech.
Proprietary training methodology, model architecture, and training data to improve baseline accuracy.
Custom SLMs trained on your corpus, your speakers, your terminology.
On-prem deployment and custom agent architectures for enterprise-grade security.
Lightweight models designed to run on CPUs for maximum accessibility.
End to end platform for agent orchestration, with custom logic for enterprise workflows.
Commitment to open models for the community on Hugging Face.

Arti Khanijo
Chief Product Officer

12
patents filed
23
world records
18+
peer-reviewed publications
216+
languages supported
Zero STT:
3.10% word error rate
best in the industry.
Voice support for orders, returns, payments, and customer queries.
Smarter conversations with transcription, sentiment, and agent intelligence.
Secure voice AI for KYC, fraud detection, collections, and compliance.
Clinical documentation, transcription, and patient conversations that capture every detail.
Automate customer support, retention, upgrades, and service interactions.
In-store voice assistance for product queries, inventory, and POS support.
We're building voice infrastructure for the next decade.
Hiring engineers, researchers and linguists who care about deployable intelligence.