How software designers are grappling with creating a computer voice that people like, which is as much an art as it is a science
Creating a Computer Voice That People Like — When computers speak, how human should they sound? — This was a question that a team of six IBM linguists …
Context & Ripple Effects
When a team of six IBM linguists sat down to answer how human a computer voice should sound, voice design was an in-house craft problem — equal parts art and science, solved by small specialist teams inside big companies. The question they framed became the founding brief for the next decade of interface work, from developers struggling to humanize virtual assistant voices in the years after this piece ran to researchers making bots genuinely conversational.
The arc since has run in one direction: what IBM's linguists treated as a bespoke design exercise is now a market. Startups such as WellSaid Labs and VocaliD sell custom AI voice actors for assistants, game characters, and corporate video, while Amazon's researchers attack the underlying engineering problem of voice disentanglement to make devices more conversational.
First-order effects
- Voice design shifts from an internal IBM linguistics team's judgment call to a defined discipline, forcing every company shipping a speaking product to decide explicitly how human its voice should be rather than defaulting to whatever the speech engine produces.
Second-order effects
- A supplier layer forms around that decision: dedicated voice-actor startups let any developer buy a distinctive persona instead of building one, turning voice quality into a competitive differentiator for assistants and games.
- As voices get more convincing, the human-likeness choice stops being cosmetic — Amazon-style work on conversationality raises the ceiling on what users expect every spoken interface to do.
Third-order effects
- If voices keep approaching human fidelity, the trust question surfaces as the binding constraint: HCI experts already warn that humanlike chatbots create cognitive dissonance about how much to believe them, which points toward norms or rules governing how machine voices may present themselves.
- The longer pattern runs toward ambient computing where the voice is the primary interface — meaning the aesthetic judgments of small design teams become infrastructure decisions affecting how hundreds of millions of people relate to software.
The trend: Computer-voice design is maturing from a niche linguistic craft into a commercialized discipline whose central trade-off — how human to sound versus how much trust that buys — is becoming a governance question.