The adoption of voice technologies by hundreds of millions of users is a positive development, but highlights the fact that speech recognition does not work equally well for some segments of society, notably children, seniors, and accented speakers. These users may become alienated by poorly performing speech technology and cut off from its benefits. This presentation characterizes the extent of the problem, summarize research in applying speaker adaptation and deep learning to address the problem, and makes a case for preventing a discriminatory “speech divide” via more inclusive speech technology.