The User Experience of today's Voice AI systems is often just mediocre. One of the biggest hurdles is that even the most advanced synthetic voices don't sound right in context. The non-verbal signals (intonation, rhythm, stress patterns etc.) they send apart from words tend to be awkward or meaningless. To make things worse, they ignore most if not all non-verbal signals conveyed by users. As a result, users feel like the system does not really listen.

Further typical user impressions of synthetic voices include:

• Lack of intelligence and competence
• Lack of politeness
• Lack of sincerity
• Lack of empathy / "fake" empathy
• Wrong or unclear signs of irony or humor
• Exaggerated / inappropriate confidence
• Lack of social accommodation
• Lack of trust, connection, and familiarity
• Listening strain (especially for elderly, hearing impaired, or non-native people, but even for native speakers with normal hearing)

Virtual personal assistants tend to get away with these issues because people use them out of personal choice and for convenience or fun. The same is not true for applications in emotionally or socially complex settings like healthcare, elder care, insurance claims, customer support, or public services – especially when users have no choice but to interact with the system.