Shunya Labs Releases Zero Codeswitch Speech Model
Shunya Labs announced Zero Codeswitch, a speech-recognition foundation model that understands naturally code-mixed and multilingual Indian speech, claiming a 3.10% WER on the OpenASR leaderboard and a 48% improvement over the next-best model. The company says the model runs on standard CPUs with sub-100 ms latency, can reduce deployment costs up to 20×, supports on-premises and air-gapped use, and meets HIPAA, SOC 2 Type II and ISO 27001 standards.
Key Points
- 1Achieves 3.10% WER on OpenASR leaderboard, 48% better than nearest competing model
- 2Trained on millions of hours of Indian code-mixed speech, handling Hinglish and regional dialects
- 3Enables CPU-first, on-premises deployments with sub-100ms latency, reducing deployment costs up to 20×
Scoring Rationale
Strong novelty and deployability from 48% WER improvement and CPU-first design, but claims rely on a company announcement.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
