SELMA: A Speech-Enabled Language Model for Virtual Assistant Interactions