[Feature]

[nwhite872]

Summary

Gemma4 E2B and E4B are capable of audio transcription, so it would be more efficient to pass audio directly to those models.

Problem / Motivation

This would reduce latency for those.

Proposed Solution

Voice mode with an audio input capable model could pass the audio to the model as an attachment and bypass the whisper transcription process.

This might improve latency and reduce ram utilization.

Platform

• Android
• iOS
• Both