There may be a gap between academic benchmarks (such as Big Bench Audio) and real-world call performance in noisy environments, and the actual deployment effectiveness of AI voice products requires more user case validation.