hexgrad/Kokoro-82M support

[umuttalha]

Problem description

react-native-executorch provides excellent on-device support for LLM, Vision, and Speech-to-Text models. However, it currently lacks a Text-to-Speech (TTS) model. This prevents developers from creating fully offline, voice-interactive experiences

Proposed solution

I propose the integration of hexgrad/Kokoro-82M, a high-quality, lightweight (82M parameters) open-source TTS model perfect for mobile devices.

Alternative solutions

No response

Benefits to React Native ExecuTorch

Complete Conversational AI: Enables a full on-device loop: Speech-to-Text -> LLM -> Text-to-Speech.
New App Capabilities: Unlocks the potential for offline voice assistants, accessibility tools, and educational apps.
Improved Developer Experience: Provides a simple, declarative API for a powerful feature without relying on external services.

Additional context

No response

[jakmro]

Hi @umuttalha, thanks for your feature proposal. We are aware that TTS support is currently missing in our library. We tried to export the hexgrad/Kokoro-82M, but encountered issues related to ExecuTorch itself. Once those are resolved, we will for sure add support for it.

[umuttalha]

Hi @jakmro,

Thanks for your response and for considering the integration of Kokoro-82M for TTS support in react-native-executorch. I’m excited about the potential this feature brings to the ecosystem.
Could you share any estimated timeline or roadmap for when the ExecuTorch-related issues might be resolved and TTS support could be added? This would help me plan my project accordingly.

Thanks in advance for any insights!

[jakmro]

@umuttalha, this is the issue that needs to be resolved, unfortunately I don't know when this will happen