Your voice.
Your phone.
No subscription.

Speech

Made for listening, made for you.

Once the required files are imported, speech generation, saved voices and recordings work without an internet connection. Internet access is used only when you request an app update in Settings. Speech, voice and recording data stay on your device; no automatic update polling runs. Downloads and this website need a connection.

28 October 2026 / Android code 19

New in prototype 0.1.18

Balanced reuses prepared voice information between passages while preserving the model, precision, voice and delivery. One bounded cache uses about 26.4 MiB of extra memory and is released on voice change or engine close; cancelled preparation cleans up before retry. Three Windows CPU comparisons measured generation speedups of 1.12–1.17x, with identical tokens and audio samples for those fixtures. Total time including preparation fell from 24.44 to 22.04 seconds. These results are not a phone-speed, all-text or listening-quality guarantee; phone performance, memory and listening checks have not yet been performed. FP32, local models/voices, recordings, settings and earlier buffering/background controls are preserved.

New in prototype 0.1.17

While generated audio waits for the smooth-playback reserve, the app shows Buffering audio and Play available audio. Tap Play available audio to play the ready portion immediately. Early playback can have gaps when synthesis is slower; the default automatic reserve remains. An explicit Play/Resume choice stays enabled for later parts and resets on Stop or a new task. Playback does not start while the app is hidden. Models, voices, recordings, settings and background generation are preserved. Host checks confirm non-silent generated audio, but do not prove audible output on your phone. Phone listening, volume and audio-output checks have not yet been performed.

New in prototype 0.1.16

User-started speech generation, voice previews and comparisons can continue while switching apps or the screen sleeps. The current job and progress survive Activity recreation. Playback pauses while hidden and resumes on request. Optional progress notifications and Stop can be enabled in Settings; denying notifications does not cancel the job, which Android may show in its task manager instead. Android time limits, force-stop, reboot or process death can still interrupt work; automatic restart is not guaranteed. Imports and microphone capture stay in the foreground. Models, voices, recordings and settings are preserved. Phone tests of background work, screen-off, recreation, Stop and listening have not yet been performed.

New in prototype 0.1.15

Use Check for updates in Settings to download a newer prototype APK from the official Speech page. The app checks package, version, prototype channel and signing identity before Android asks for installation permission and consent. No silent installation, downgrade or data clearing. Internet is used only for this requested update; text, voices, recordings and speech generation stay local. Includes the performance and recording changes from 0.1.13 and 0.1.14. Phone tests of networking, permissions and installation have not yet been performed.

New in prototype 0.1.14

Saved and recent recordings show the actual WAV bitrate and file size. Hold a recent recording to choose Delete or Cancel, including the last recording. Confirmation removes only that private recording and its metadata; exported files, other recordings, models, voices and settings are preserved. Playback of the deleted recording stops. Includes the performance improvements from 0.1.13. Phone tests of recording details, long press, confirmation and playback have not yet been performed.

New in prototype 0.1.13

The model prepares in the background. Saved voices switch without reloading Balanced model sessions. Adaptive audio buffering prepares later parts during playback. Slower phones or text can need longer initial buffering; uninterrupted playback is not guaranteed. Models, voices, settings and recordings are preserved. Phone listening and lifecycle tests have not yet been performed.

New in prototype 0.1.12

A more compact Speech and Voices interface with clearer actions and shorter guidance. Version, model import and setup tools are in Settings. Existing models, voices, drafts and recordings are preserved.