Scaling AI voices to 20+ locales

01 / THE OPPORTUNITY

Studio production limited how many locales Waze could serve.

Waze’s ambition to bring locally relevant voice experiences to more audiences ran into the cost and time of traditional production. Generative AI offered a way to explore global scale within a limited budget.

03 / THE CREATIVE OPPORTUNITY

Helped voice talent reach new audiences in new languages.

I saw a promising use for generative AI: enabling voice actors to speak in languages Waze had not previously been able to support. I found new and existing talent open to experimenting.

We already had voice campaigns we could build on. To keep costs down, we reviewed which campaigns would resonate in each market, then worked with a transcreation agency to adapt the selected scripts for local audiences.

04 / THE PARTNERSHIPS & SAFEGUARDS

Built talent protections into the AI workflow.

Introducing AI voice generation required rigorous cross-functional alignment. I collaborated directly with DeepMind, Product, and Engineering to adapt our existing legal agreements and establish a clear ethical baseline.

I negotiated strict data protections to ensure our actors’ voices would never be used to train future models. This critical safeguard helped give talent and internal creative teams the confidence to participate.

05 / THE QUALITY BAR

Created a four-part quality standard for AI voices.

I created a scoring system to help decide which generated prompts were suitable to go into the app. It assessed:

• Consistency of voice
• Cultural fit
• Functional clarity
• Character believability

06 / FROM MANUAL TO SELF-SERVE

From four days to a few hours.

Listeners accepted AI voices at the same rate as studio recordings, giving us confidence to scale. We moved from days of manual notebook work to AI Studio, where creative leads can generate voices independently in hours.

Reusable voice models reduced studio and re-licensing costs, saving hundreds of thousands of dollars across 20 World Cup languages alone.