Dictate

一款轻量级的 Android 语音输入工具，可与你现有的键盘配合使用，支持可配置的音频编码、兼容 OpenAI 的转录 API，以及剪贴板回退。

Dictate is an Android voice-transcription enhancement layer, not an input method editor. It shows one non-focusable accessibility overlay, records one voice task, sends it directly to the user's OpenAI-compatible endpoint, and inserts the returned top-level text into the editable focus that exists when the response arrives.

Features No IME, keyboard, candidate bar, editing context, history list, cloud account, or proxy server. AudioRecord PCM capture at the active input route's selected sample rate, in 16-bit mono, with pause/resume, microphone foreground service, wake lock, and cancellation. Draggable, non-focusable accessibility overlay with persisted, inset-aware screen position; its size, opacity, and recording/paused/processing color scheme are configurable and update without restarting the accessibility service. An in-progress recording continues while the screen is off; no lock-screen controls or lock-screen text insertion are provided. Embedded FFmpeg n8.1, Opus 1.5.2, and LAME 3.100, built from source for arm64-v8a only. Opus, MP3, AAC, and PCM/WAV output with valid codec/container choices only; output rate follows the active capture rate by default (capped at 48 kHz). Direct multipart requests to OpenAI-compatible /v1/audio/transcriptions; success requires a non-empty top-level JSON string text. Current-focus insertion plus a configurable clipboard safety copy, enabled by default; when disabled, clipboard is used only after an explicit insertion failure. Cancellable FFmpeg process, HTTP request, and exponential retry wait, all protected by a monotonically increasing task ID. Keystore-backed API Key encryption, redacted diagnostics, real endpoint test, and validated JSON import/export.