Are there any associated benchmarks with this? Like how accurate is this technology for different types of facial structures, sentences or languages?