A deep joint sparse non-negative matrix factorization framework for identifying the common and subject-specific functional units of tongue motion during speech.