My Framer Site

Designing voice ux for in-car assistant system

defining voice identity and conversation design framework

Timeline

Jan - Jun 2024

Role

Product Designer

Live Product

TataCurvvEV

Team

Team of 3

The Brief

How should an in-car voice assistant sound, behave, and respond across driving contexts?

The client wanted to answer a more fundamental question, what should this in-car voice actually be? They wanted something that felt strongly connected to the brand, competitive in the market, and more human-centred than existing assistants, going beyond just subjective preferences.

Outcome

A research-backed voice identity and a scalable semantic framework

The outcome was a repeatable system for making voice decisions that the client could align teams around and extend over time. Voice shifted from a subjective layer to a designed system with clear logic.

Elements In Focus

Voice Quality

voice accent, perceived age, speech rate, tonality

Brand Connect

how well a voice embodied the client's intended positioning

User Connect

likability score and user role adaptability score

Phase 1

Structuring a voice criteria

Extensive secondary research established 3 elements (mentioned above) voice should be evaluated on. To test voices based on these elements, research instruments like a Voice Persona, a Wizard of Oz testing guide, an In-Cabin testing guide, a Likability Score metric, and a Voice Role Validation guide were formed.

Phase 2

User testing and voice finalisation

Based on persona, professional voice recordings were sourced. Voice samples were paired with contextual prompts and visuals to reflect realistic in-car usage rather than abstract listening conditions. Post user testing, total voice score for each sample gave a clear winner voice.

Phase 3

Designing how voice behaves

A semantic framework was designed which showed how the voice assistant's role adapted based on the user's intent. Three roles were defined - Assistant, Coach and Companion, derived from existing literature. Every voice command was assigned an intent, every intent was given a voice role.

Process Pivots

Some important questions came up and pivots happened in the process which we had to answer for, like -

Outcome

The voice assistant launched in the Tata Curvv in 2024.

Apart from being recognised the most creative and thoughtfully structured efforts of the year within Tata Elxsi, the framework also became a foundation for future voice UX work internally.

Reflection

A shift towards a methodology and framework-first thinking.

The absence of a traditional final output, no screen, no interface, no flow was initially unfamiliar. What emerged instead was a methodology and a framework, and learning to treat process and logic as the product was one of the most important shifts this project produced. I’m happy to chat more about my process over a call. Reach out to me at bhavikamalik@u.nus.edu !