Many apps need voice features but can't use cloud services due to data residency rules or privacy requirements.
Package local speech models into easy-to-integrate modules for common development frameworks, with compact models that run on-device.
Healthcare and legal tech developers would pay for compliant voice integration kits.
Start with basic TTS that works offline and maintains good-enough quality for most use cases.
The risk is cloud providers lowering costs enough to remove the local advantage.