A set of therapy parameter values is selected based on a patient state, where the patient state comprises a speech state or a mixed patient state including the speech state and at least one of a movement state or a sleep state. In this way, therapy delivery is tailored to the patient state, which may include one or more patient symptoms specific to the patient state. In some examples, a medical device determines whether the patient is in the speech state or a mixed patient state including the speech state based on a signal generated by a voice activity sensor. The voice activity sensor detects the use of the patient's voice, and may include a microphone, a vibration detector or an accelerometer.