Synthesized speech evals generated by MistralAI from popular text evaluation datasets to evaluate spoken-language reasoning capabilities of Audio LLMs