A new USC study finds that software detects changes in clinical states in voice data for patients with bipolar, schizophrenia and depressive disorders as accurately as attending doctors.