Humyn Labs Releases BRIDGE Benchmark for Voice AI in Global South Languages

Humyn Labs has published BRIDGE, a large independent benchmark for commercial AI speech recognition tools covering 22 non-English languages across the Global South, announced in a press release. The benchmark evaluates 15 models using real conversational data from languages spoken by more than 5.5 billion people.

The report shows that global performance rankings do not always reflect regional accuracy. ElevenLabs Scribe v2 achieved the lowest overall word error rate at 10.6 percent but ranked third in Vietnamese, while AssemblyAI Universal performed best in that language despite ranking lower globally. Significant performance gaps were also observed in Venezuelan Spanish, Brazilian Portuguese, and cross-regional speaker tests.

BRIDGE uses seven evaluation metrics beyond standard word and character error rates, such as Semantic Similarity, Code Switch F1, Loan Word WER, Phoneme Informed Error Rate, and Word Information Lost. The dataset was built on unscripted two-person conversations collected from multiple regions, and the full dataset is available on Hugging Face.