Releases: Today20092/voice-input
Release list
FUTO Voice Input Moonshine v1.4.2 — S1-mini
S1-mini stable release
- Adds optional on-device English transcript cleanup with S1-mini by Superwhisper.
- Keeps Moonshine v2 Small Streaming as the default offline recognizer.
- Reorganizes Model Options with behavior, languages, download size, installed size, status, and model details.
- Keeps Moonshine Medium, Parakeet Unified, Parakeet TDT, Nemotron, and legacy Whisper/GGML selectable.
- Adds S1-mini style, structure, context, keep-warm, CPU/OpenCL optimization, and diagnostics controls.
- Includes benchmark validation and Android native-backend discovery fixes from the beta cycle.
Installation and models
This is the tested, signed standalone APK previously published as beta 11. The APK is promoted unchanged so it matches the validated build and SHA-256 digest.
Speech models and S1-mini weights are not bundled. Download the desired model on first use or from settings; transcription and cleanup remain offline after installation. The first supported ABI is arm64-v8a.
S1-mini is disabled by default. If cleanup times out, fails, encounters memory pressure, or cannot establish English, the app safely returns the raw transcript. OpenCL depends on the phone vendor's driver and falls back to validated CPU inference.
Standard diagnostics never include audio, transcripts, prompts, or vocabulary. Transcript capture is a separate opt-in feature with explicit export labeling and retention controls.
Full changelog: v1.4.1...v1.4.2-beta.11
FUTO Voice Input Moonshine v1.4.2-beta.9
Beta 9 highlights
- Extracts packaged llama.cpp CPU and OpenCL backend libraries so Android's filesystem loader can discover them.
- Prevents release publication unless the APK manifest requires native-library extraction.
- Adds per-candidate benchmark failures, packaged libraries, and discovered devices to standard diagnostics.
- Adds persistent opt-in transcript capture for eligible English S1-mini cleanup runs.
- Keeps standard diagnostics transcript-free and adds a separately confirmed, clearly labeled transcript ZIP.
- Adds transcript review, individual deletion, clear-all, ten-run retention, and seven-day expiry.
What to test
- Rerun the optimization test on a Galaxy S24+; it should select a valid CPU or OpenCL result.
- Dictate English, press Stop, and confirm S1-mini changes the final transcript instead of falling back immediately.
- Enable transcript capture, perform a cleanup run, and review all three labeled transcript stages.
- Export both ZIP types and confirm only the
WITH-TRANSCRIPTSarchive contains dictated text. - Attach the new standard or transcript-inclusive ZIP with any failure or quality feedback.
Known beta gaps
- The backend-loader fix still needs validation on the target Galaxy S24+.
- OpenCL support depends on the phone vendor's driver and automatically falls back to validated CPU inference.
- Cleanup may return the raw transcript after a timeout, model/process failure, or memory pressure.
- Release-level thermal, memory, long-transcript, and full backend regression testing remains in progress.
Installation and models
This is a signed standalone APK for the FUTO Voice Input fork. Moonshine remains the default offline recognizer; Parakeet, Nemotron, and legacy Whisper/GGML remain selectable in Model Options.
Speech models and S1-mini weights are not bundled in the APK. Download the desired model on first use or from Model Options; transcription and cleanup remain offline after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.8...v1.4.2-beta.9
FUTO Voice Input Moonshine v1.4.2-beta.8
Beta 8 highlights
- Fixes Android discovery of the packaged llama.cpp CPU and OpenCL backend libraries.
- Adds per-candidate benchmark failures, packaged libraries, and discovered devices to standard diagnostics.
- Adds persistent opt-in transcript capture for eligible English S1-mini cleanup runs.
- Keeps standard diagnostics transcript-free and adds a separately confirmed, clearly labeled transcript ZIP.
- Adds transcript review, individual deletion, clear-all, ten-run retention, and seven-day expiry.
What to test
- Rerun the optimization test on a Galaxy S24+; it should select a valid CPU or OpenCL result.
- Dictate English, press Stop, and confirm S1-mini changes the final transcript instead of falling back immediately.
- Enable transcript capture, perform a cleanup run, and review all three labeled transcript stages.
- Export both ZIP types and confirm only the
WITH-TRANSCRIPTSarchive contains dictated text. - Attach the new standard or transcript-inclusive ZIP with any failure or quality feedback.
Known beta gaps
- The backend-loader fix still needs validation on the target Galaxy S24+.
- OpenCL support depends on the phone vendor's driver and automatically falls back to validated CPU inference.
- Cleanup may return the raw transcript after a timeout, model/process failure, or memory pressure.
- Release-level thermal, memory, long-transcript, and full backend regression testing remains in progress.
Installation and models
This is a signed standalone APK for the FUTO Voice Input fork. Moonshine remains the default offline recognizer; Parakeet, Nemotron, and legacy Whisper/GGML remain selectable in Model Options.
Speech models and S1-mini weights are not bundled in the APK. Download the desired model on first use or from Model Options; transcription and cleanup remain offline after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.7...v1.4.2-beta.8
FUTO Voice Input Moonshine v1.4.2-beta.7
Beta 7 highlights
- Adds optional on-device final-transcript cleanup with S1-mini by Superwhisper.
- Clearly labels the feature English-only and bypasses cleanup unless English is established.
- Adds styling, structure, context, model warm-time, and runtime controls.
- Downloads and verifies the pinned Q4_K_M model at runtime; model weights are not bundled in the APK.
- Benchmarks CPU thread counts and experimental OpenCL on the phone, selecting OpenCL only when its validated result is at least 15% faster.
- Runs cleanup in an isolated app process with a 15-second raw-transcript fallback.
- Adds transcript-free performance diagnostics with copy and ZIP export.
What to test
- Download S1-mini from Model Options, enable it, and dictate English on a Galaxy S24+.
- Try every styling, structure, and context option and confirm cleanup runs only after Stop.
- Try non-English and mixed-language configurations; confirm raw text is retained when English is not established.
- Compare Auto, CPU, and OpenCL and rerun the optimization test while watching latency and thermals.
- Test immediate unload and each keep-warm duration, cancellation, long dictation, and low-memory recovery.
- Export the diagnostics ZIP after any slow or incorrect run and attach it with feedback.
Known beta gaps
- This beta has not yet been performance-validated on the target Galaxy S24+.
- OpenCL support depends on the phone vendor's driver and automatically falls back to validated CPU inference.
- Cleanup may return the raw transcript after a timeout, model/process failure, or memory pressure.
- Release-level thermal, memory, long-transcript, and full backend regression testing remains in progress.
Installation and models
This is a signed standalone APK for the FUTO Voice Input fork. Moonshine remains the default offline recognizer; Parakeet, Nemotron, and legacy Whisper/GGML remain selectable in Model Options.
Speech models and S1-mini weights are not bundled in the APK. Download the desired model on first use or from Model Options; transcription and cleanup remain offline after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.6...v1.4.2-beta.7
FUTO Voice Input Moonshine v1.4.2-beta.10
Beta 10 highlights
- Fixes the optimization test rejecting valid S1-mini rewrites that differ from one exact sentence.
- Validates three stable outputs for required meaning before timing a backend.
- Skips OpenCL benchmarking when Android discovers no GPU backend.
- Records validation hashes, stability, meaning checks, and skipped candidates in diagnostics.
- Extracts packaged llama.cpp CPU and OpenCL backend libraries so Android's filesystem loader can discover them.
- Prevents release publication unless the APK manifest requires native-library extraction.
- Adds per-candidate benchmark failures, packaged libraries, and discovered devices to standard diagnostics.
- Adds persistent opt-in transcript capture for eligible English S1-mini cleanup runs.
- Keeps standard diagnostics transcript-free and adds a separately confirmed, clearly labeled transcript ZIP.
- Adds transcript review, individual deletion, clear-all, ten-run retention, and seven-day expiry.
What to test
- Rerun the optimization test on a Galaxy S24+; it should select a valid CPU or OpenCL result.
- Dictate English, press Stop, and confirm S1-mini changes the final transcript instead of falling back immediately.
- Enable transcript capture, perform a cleanup run, and review all three labeled transcript stages.
- Export both ZIP types and confirm only the
WITH-TRANSCRIPTSarchive contains dictated text. - Attach the new standard or transcript-inclusive ZIP with any failure or quality feedback.
Known beta gaps
- The backend-loader fix still needs validation on the target Galaxy S24+.
- OpenCL support depends on the phone vendor's driver and automatically falls back to validated CPU inference.
- Cleanup may return the raw transcript after a timeout, model/process failure, or memory pressure.
- Release-level thermal, memory, long-transcript, and full backend regression testing remains in progress.
Installation and models
This is a signed standalone APK for the FUTO Voice Input fork. Moonshine remains the default offline recognizer; Parakeet, Nemotron, and legacy Whisper/GGML remain selectable in Model Options.
Speech models and S1-mini weights are not bundled in the APK. Download the desired model on first use or from Model Options; transcription and cleanup remain offline after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.9...v1.4.2-beta.10
FUTO Voice Input Moonshine v1.4.2-beta.6
Beta 6 highlights
- Centralizes recognition-model readiness, loading, selection, installation, deletion, and runtime release.
- Keeps recording active while streaming models load so early speech is not dropped.
- Preserves queued streaming audio when decoding falls behind and rejects stale samples after a recording session ends.
- Removes full model-file hashing from interactive startup while retaining integrity checks during installation.
- Recovers more safely from model-load failures and now closes partially initialized backends when loading fails or is cancelled.
- Adds predictive-back transitions in Settings on supported Android versions.
- Expands regression coverage for recording stop policy, streaming replay/reset, catalog readiness, and model lifecycle behavior.
What to test
- Start speaking immediately after opening voice input, especially with a streaming model selected.
- Cancel and rapidly restart dictation; confirm no text or audio leaks between sessions.
- Switch among installed Moonshine, Parakeet, Nemotron, and Whisper models and verify selection survives restart.
- Interrupt or fail a model download/load and confirm the app offers a clean retry without deleting a valid download.
- Exercise Android 15 predictive back from Model Options and report animation or back-stack issues.
Known beta gaps
- Predictive-back animation still needs broader real-device and instrumentation validation.
- Model performance labels and latency guidance still need calibration across representative phones.
- Release-level thermal, memory, backlog, download-recovery, and full backend regression testing remains in progress.
Installation and models
This is a signed standalone APK for the FUTO Voice Input fork. Moonshine remains the default offline recognizer; Parakeet, Nemotron, and legacy Whisper/GGML remain selectable in Model Options.
Speech models are not bundled in the APK. Download the desired model on first use or from Model Options; transcription remains offline after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.5...v1.4.2-beta.6
FUTO Voice Input Moonshine v1.4.2-beta.5
Beta 5 highlights
- Centralizes recognition-model readiness, loading, selection, installation, deletion, and runtime release.
- Keeps recording active while streaming models load so early speech is not dropped.
- Preserves queued streaming audio when decoding falls behind and rejects stale samples after a recording session ends.
- Removes full model-file hashing from interactive startup while retaining integrity checks during installation.
- Recovers more safely from model-load failures and now closes partially initialized backends when loading fails or is cancelled.
- Adds predictive-back transitions in Settings on supported Android versions.
- Expands regression coverage for recording stop policy, streaming replay/reset, catalog readiness, and model lifecycle behavior.
What to test
- Start speaking immediately after opening voice input, especially with a streaming model selected.
- Cancel and rapidly restart dictation; confirm no text or audio leaks between sessions.
- Switch among installed Moonshine, Parakeet, Nemotron, and Whisper models and verify selection survives restart.
- Interrupt or fail a model download/load and confirm the app offers a clean retry without deleting a valid download.
- Exercise Android 15 predictive back from Model Options and report animation or back-stack issues.
Known beta gaps
- Predictive-back animation still needs broader real-device and instrumentation validation.
- Model performance labels and latency guidance still need calibration across representative phones.
- Release-level thermal, memory, backlog, download-recovery, and full backend regression testing remains in progress.
Installation and models
This is a signed standalone APK for the FUTO Voice Input fork. Moonshine remains the default offline recognizer; Parakeet, Nemotron, and legacy Whisper/GGML remain selectable in Model Options.
Speech models are not bundled in the APK. Download the desired model on first use or from Model Options; transcription remains offline after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.4...v1.4.2-beta.5
FUTO Voice Input Moonshine v1.4.2-beta.4
Offline Moonshine streaming release of the FUTO Voice Input fork, with Parakeet and Whisper retained as selectable backends.
Models download on first use or from Model Options, keeping the APK smaller while preserving offline transcription after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.3...v1.4.2-beta.4
FUTO Voice Input Moonshine v1.4.2-beta.3
Offline Moonshine streaming release of the FUTO Voice Input fork, with Parakeet and Whisper retained as selectable backends.
Models download on first use or from Model Options, keeping the APK smaller while preserving offline transcription after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.2...v1.4.2-beta.3
FUTO Voice Input Moonshine v1.4.2-beta.2
Offline Moonshine streaming release of the FUTO Voice Input fork, with Parakeet and Whisper retained as selectable backends.
Models download on first use or from Model Options, keeping the APK smaller while preserving offline transcription after installation.
The Parakeet integration and repository changes in this fork were built with AI assistance using Codex (GPT-5).
Full Changelog: v1.4.2-beta.1...v1.4.2-beta.2