We present an Emotional Music Information Retrieval system for mobile devices that utilizes a machine learning approach to detect latent emotion from within both user queries (non-descriptive queries) and the lyrics of songs and uses both elements to develop an effective Music Information Retrieval system. Emotion is extracted from the songs and queries and mapped into a high-dimensional emotion space, which allows for the employment of conventional text retrieval techniques to calculate the similarity between a user query and the latent emotion in song lyrics, thereby producing a ranked list of songs for playback.