News & Founder Notes
Official announcements, foundation model architecture drops, and studio insights from Darsh Yadav.
Sorika Labs Releases Vaani Speech AI Series: Zero-to-SOTA on Community GPUs
Sorika Labs officially unveils the Vaani Speech AI Series — foundation speech models engineered and converged entirely on community GPUs (Google Colab T4 & Kaggle Dual-GPU). Delivering 100% permanently fused safetensors, Vaani captures #1 SOTA on Indian Speech (11.78% Hindi WER) in the Lite Series, and #1 English Accuracy (5.90% WER) with 21.5x–30.3x Real-Time velocity in the Flash Series.
Sorika Labs Releases Vaani Speech AI Series: Zero-to-SOTA on Community GPUs
Sorika Labs officially unveils the Vaani Speech AI Series — foundation speech models engineered and converged entirely on community GPUs (Google Colab T4 & Kaggle Dual-GPU). Delivering 100% permanently fused safetensors, Vaani captures #1 SOTA on Indian Speech (11.78% Hindi WER) in the Lite Series, and #1 English Accuracy (5.90% WER) with 21.5x–30.3x Real-Time velocity in the Flash Series.
Sorika Labs Accelerates Active R&D on Swen Series, Vision, Nano & Fast Thinking Models
Sorika Labs is actively advancing Research & Development across the upcoming Swen Series, next-generation Vision models, ultra-compact Nano models, and high-throughput Thinking Fast reasoning engines — slated for official public release within the next 2 months.
Meet Reverie: The Conversational Intelligence & Rapid Prototyping Workspace by Sorika
Sorika AI Labs is officially introducing Reverie — our flagship conversational AI workspace. Engineered like DeepSeek for nuanced dialogue, live web search, precision tool calling, interactive diagrams, and lightweight website and game prototyping.
Introducing Sorika AI Labs: An Independent Studio for Compact Foundation Models
Founded by Darsh Yadav, Sorika AI Labs is launching to build lightweight, sovereign, and deterministic foundation models across speech, vision, and reasoning.
Introducing Vaani 1 Lite: Streaming Multilingual Speech Recognition for Indic Dialects
Vaani 1 Lite introduces a compact conformer-transducer foundation model for streaming speech recognition across 22 scheduled Indian languages with sub-100ms latency.
Introducing Suzune S1: Ultra-Fast 80M Parameter Non-Autoregressive Speech Foundation Model
Suzune S1 is an ultra-lightweight 80M parameter speech synthesis engine delivering 24kHz broadcast audio with sub-20ms streaming latency and constant 320MB memory.