High-quality, multi-environment, speech data for high performance language models