Build an Android voice-input method that works beside a user's existing keyboard and saves audio locally before transcription begins. It should provide live text, offline recognition, configurable punctuation commands, pause tolerance, and automatic recovery or retry without losing the recording. Start with a narrow paid product for dependable cross-app dictation rather than another feature-heavy keyboard.

Recent reviews across Gboard, FUTO Keyboard, Dictate Keyboard, Yandex Keyboard, SwiftKey, and other products show persistent failures in the same workflow. Capable on-device speech models now make privacy-preserving transcription and recovery possible without depending entirely on a network service.