iFLYTEK ships Spark-ASR-2.0 for noise, dialects and code-switch

iFLYTEK released Spark-ASR-2.0, its latest speech-recognition model, on September 23, 2026, targeting noisy, dialect, low-volume and Chinese-English code-switch scenes. The company claims lower word-error rates than Spark-ASR-1.0 with only about 10% higher inference cost. Rollout starts with iFLYTEK Input on September 24, plus an open API.

• Claims lower error rates in noise, dialects and code-switch
• Inference cost rose only about 10%
• Dialect coverage spans 202 Chinese cities
• Rollout into iFLYTEK Input and open API from September 24