FUTO Voice Input Moonshine

This personal fork keeps the FUTO voice keyboard experience, uses Moonshine v2 Streaming as its default offline recognizer, and adds optional on-device transcript cleanup with S1-mini by Superwhisper. Moonshine Small and Medium, Parakeet, Nemotron, and legacy FUTO Whisper/GGML models are available from Model Options.

This fork's Parakeet integration and repository changes were built with AI assistance using Codex (GPT-5).

The goal is straightforward: keep the FUTO UI and recording flow while adding responsive streaming transcription, personal vocabulary corrections, and private offline transcript cleanup.

What Changed

• Moonshine v2 Small Streaming is the default balanced option and emits live partial transcripts.
• Moonshine v2 Medium Streaming is available as a higher-accuracy option.
• Parakeet, Nemotron, and legacy Whisper/GGML remain selectable backends.
• Batch and streaming recognizers share backend-neutral Kotlin contracts.
• Personal vocabulary entries correct partial and final transcripts; use heard => preferred for explicit aliases.
• Optional S1-mini by Superwhisper cleanup runs locally on final English transcripts after Stop.
• S1-mini includes styling, structure, context, keep-warm, CPU/OpenCL, optimization, and transcript-free diagnostics controls.
• The stable app uses the distinct org.futo.voiceinput.moonshine package ID.
• Only the selected backend's model files are required before voice input starts.

Screenshots

Model Options

Moonshine is selected by default, with Parakeet, Nemotron, and Whisper/GGML available as alternatives.

Project branches

• master contains the current stable release.
• codex/s1-mini-beta preserves the S1-mini development and release-validation history.

Active Model

The default backend and model are:

Moonshine v2 Small Streaming English

The app downloads the selected quantized model assets from:

https://download.moonshine.ai/model/small-streaming-en/quantized/
https://download.moonshine.ai/model/medium-streaming-en/quantized/

Model files are downloaded on first use rather than packaged into the APK.

After installing the APK, the model is downloaded by the app:

1. Open FUTO Voice Input Moonshine Settings.
2. Tap Model.
3. Select Balanced (Small) or Higher accuracy (Medium).
4. Confirm the download.

If you try voice input before downloading the model, the app prompts for the download. Transcription runs offline after installation.

Downloaded model files are stored in app-private storage:

filesDir/moonshine-small-streaming-en/
filesDir/moonshine-medium-streaming-en/

Parakeet and Whisper/GGML models are also stored in app-private storage and run offline after download.

Optional S1-mini transcript cleanup

Open Transcript Cleanup from the main settings screen to download and enable S1-mini by Superwhisper. The pinned Q4_K_M model is approximately 484.2 MB. It is English-only, disabled by default, and runs only after recording stops. If cleanup times out, fails, or English is not established, the app keeps the raw transcript. Personal Vocabulary corrections run after cleanup.

The first run benchmarks validated CPU configurations and experimental OpenCL on the phone. The Auto setting selects OpenCL only when it produces the expected output and is at least 15% faster than the best CPU result. Diagnostics contain timing, runtime, memory, and thermal data only—never audio, transcripts, prompts, or vocabulary—and can be exported as a ZIP for bug reports.

Building Locally

Required tools:

• Android Studio or Android SDK command line tools
• JDK 17 or newer
• Android SDK platform 35
• Android NDK 28.2.13676358

Create local.properties if Android Studio has not already created it:

sdk.dir=C\:\\Users\\User\\AppData\\Local\\Android\\Sdk

Build the debug APK:

.\gradlew.bat :app:assembleDevDebug

If you need a development build that packages the Parakeet model into the APK, enable the bundled-model Gradle property:

.\gradlew.bat :app:assembleDevDebug -PbundleParakeetModel=true

The APK is written to:

app/build/outputs/apk/standalone/release/app-standalone-release.apk

GitHub Releases

The current stable release is the tested v1.4.2-beta.11 build, promoted unchanged after beta validation. Download it from GitHub Releases.

This repository includes a GitHub Actions workflow that builds and verifies an APK when a v* tag is pushed.

To create a new release:

git tag v1.4.3
git push origin master
git push origin v1.4.3

GitHub Actions will:

• install Android build components
• build :app:assembleStandaloneRelease
• create a GitHub Release for the pushed tag
• attach the APK to that GitHub Release

You can also run the workflow manually from the Actions tab. Manual runs upload the APK as a workflow artifact but do not create a GitHub Release unless the run is for a tag.

Notes

• First supported ABI is arm64-v8a.
• This is intended for sideloading and personal testing.
• The normal APK does not include speech model files.
• Moonshine emits live partial transcripts; Parakeet currently returns only a final transcript.
• The app requires network access to download the selected backend's model the first time, then transcription runs offline.

Attribution And License

This fork is based on FUTO Voice Input and keeps FUTO's license and notices. FUTO Voice Input is licensed under the FUTO Source First License. Review LICENSE.md before distributing modified builds.

Parakeet model assets come from istupakov/parakeet-tdt-0.6b-v3-onnx, an ONNX conversion of NVIDIA Parakeet TDT 0.6B V3, licensed CC-BY-4.0.

The optional downloaded cleanup model is S1-mini by Superwhisper, licensed under Apache License 2.0 with the publisher's required naming condition. See the model notice. llama.cpp is included as a pinned submodule under its MIT license; Khronos OpenCL headers and loader are included as pinned submodules under their respective upstream licenses.

This fork is not affiliated with or endorsed by FUTO.