There are several applications for speech-emotion recognition (SER) systems in areas such as security and defense and healthcare. SER systems have achieved high performance when they are trained and tested in similar conditions. However, the performance often drops in more realistic and diverse conditions. Most existing SER datasets are too controlled and do not capture complex scenarios relevant to practical applications. This paper presents the White House tapes speech emotion recognition (WHiSER) corpus, which includes distant speech with real emotions from conversations in the Oval Office in 1972. This dataset is unique because it combines natural emotional expressions with various background noises, making it a perfect tool to test and improve SER models. Its real-world complexity and authenticity make the WHiSER corpus an excellent corpus for advancing emotion recognition technology, offering insights into how human emotions can be accurately recognized in complex environments.