rocky/OSGKeyboard
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: migrate on-device Qwen3 ASR to CoreML for background Flow dictation

Browse Source 
Replace MLX GPU inference with CoreML bundles so transcription continues
while the host app is backgrounded. Adds model download and warm-up,
vendored Qwen3Speech, and updates onboarding, settings, and copy for the
~1.6 GB CoreML package (iOS 18+).
This commit is contained in:
Rocky
2026-06-23 00:46:58 +08:00
parent 5e5122f172
commit df1c5ff32c
160 changed files with 22080 additions and 492 deletions
Download Patch File Download Diff File Expand all files Collapse all files
OSGKeyboard/en.lproj/Localizable.strings
+91 -4
View File
@@ -29,6 +29,7 @@
"onboarding.enable.step3.suffix" = "and select OSGKeyboard";
"onboarding.enable.openSettings" = "Open Settings";
"onboarding.api.title" = "Choose Engine";
"onboarding.api.localModels.hint" = "Download the on-device CoreML ASR model below before you finish setup (only needed when using Qwen3-ASR).";
"settings.onboarding.replay" = "Restart permission setup";
/* Common navigation */
@@ -83,15 +84,19 @@
/* Settings */
"settings.title" = "Settings";
"settings.appLanguage.title" = "App language";
"settings.appLanguage.auto" = "Auto";
"settings.appLanguage.english" = "English";
"settings.appLanguage.chinese" = "Chinese";
"settings.reset.title" = "Reset all settings?";
"settings.reset.message" = "API key, model, base URL, voice history, and cloud consent will be cleared.";
"settings.reset.confirm" = "Reset all settings";
"settings.engine.title" = "Engine";
"settings.engine.subtitle" = "Pick the recognition engine. Local engine does transcription only, no API key needed.";
"settings.engine.local.title" = "On-device";
"settings.engine.title" = "Recognition method";
"settings.engine.subtitle" = "Local: on-device ASR only. Cloud: ASR + polish via your API.";
"settings.engine.local.title" = "On-device recognition";
"settings.engine.local.ios26" = "Always on-device, no network.";
"settings.engine.local.legacy" = "On-device ASR. Transcription only, no polish.";
"settings.engine.cloud.title" = "Cloud polish";
"settings.engine.cloud.title" = "Cloud recognition & polish";
"settings.engine.cloud.subtitle" = "On-device ASR + polish via your API. Text is sent to the endpoint you configure.";
"settings.provider.title" = "Provider";
"settings.provider.subtitle" = "Pick the LLM that polishes your dictation.";
@@ -103,6 +108,12 @@
"provider.custom" = "Custom";
"settings.api.title" = "API";
"settings.language.title" = "Language";
"settings.languageModels.title" = "Language & models";
"settings.localModels.title" = "On-device models";
"settings.localModels.speechRole" = "Speech";
"settings.localModels.builtIn" = "Built-in";
"settings.localModels.allReady" = "Ready";
"settings.localModels.readiness %lld %lld" = "%lld/%lld ready";
"settings.language.subtitle.cloud" = "Recognition language and text processing mode.";
"settings.language.subtitle.local" = "Recognition language.";
"settings.mode.title" = "Mode";
@@ -110,6 +121,8 @@
"settings.mode.transcribe" = "Transcribe";
"settings.mode.polish" = "Polish";
"settings.systemPrompt.title" = "System Prompt";
"settings.systemPrompt.edit" = "Edit system prompt";
"settings.systemPrompt.hint" = "Used by the cloud LLM when polishing your dictation. Local engine skips this step.";
"settings.about.title" = "About";
"settings.systemPrompt.reset" = "Reset";
"settings.asrLocale" = "ASR locale";
@@ -128,6 +141,47 @@
"settings.privacy.cloud.alert.message" = "Cloud polish sends transcribed text to the third-party API you configure. OSGKeyboard never stores data on our servers. Continue?";
"settings.link.support" = "Help & Feedback";
"settings.link.github" = "GitHub";
"settings.support.footer" = "OSGKeyboard is open source. Report bugs and share ideas on GitHub.";
"settings.support.intro" = "The fastest way to get help or send feedback is to open a GitHub issue. We read every report.";
"settings.support.issuesHint" = "Search existing issues first. For bugs, include your iOS version, engine mode (local or cloud), and steps to reproduce.";
"settings.support.openIssues" = "Open GitHub Issues";
"settings.support.unavailable.title" = "Cannot load page";
"settings.support.unavailable.message" = "GitHub Issues could not be opened. Check your network connection and try again.";
"settings.link.licenses" = "Third-Party Licenses";
"settings.link.models" = "On-device models";
"settings.models.link.subtitle" = "Download source, models, and storage";
"settings.licenses.title" = "Third-Party Licenses";
"settings.licenses.footer" = "Open-source components used by OSGKeyboard. License texts are reproduced from upstream repositories; model weights are downloaded at runtime and cached on device. OSGKeyboard itself is source-available — commercial licensing: rocky.hk@gmail.com.";
/* On-device models manager (settings → "On-device models") */
"settings.models.title" = "On-device models";
"settings.models.intro.title" = "Run speech recognition entirely on your iPhone";
"settings.models.intro.body" = "Download the Qwen3-ASR model before using local recognition. Files are cached in this app's storage and can be deleted anytime.";
"settings.models.mirror.title" = "Download source";
"settings.models.source.modelScope" = "ModelScope";
"settings.models.source.huggingface" = "Hugging Face";
"settings.models.action.downloadSize %@" = "Download - %@";
"settings.models.action.deleteSize %@" = "Delete - %@";
"settings.models.action.downloadingPercent %lld" = "%lld%%";
"settings.models.status.notDownloaded" = "Not downloaded";
"settings.models.status.downloading" = "Downloading…";
"settings.models.status.downloadingPercent %lld" = "Downloading… %lld%%";
"settings.models.status.downloaded" = "Downloaded";
"settings.models.status.failed" = "Download failed";
"settings.models.action.download" = "Download";
"settings.models.action.cancel" = "Cancel";
"settings.models.action.cancelDownload" = "Cancel download";
"settings.models.action.delete" = "Delete";
/* Local ASR backend picker (local engine mode) */
"asr.backend.speechAnalyzer.label" = "iOS SpeechAnalyzer";
"asr.backend.speechAnalyzer.blurb" = "Built into iOS 26, always on-device, lowest latency.";
"asr.backend.speechAnalyzer.hint" = "No download required.";
"asr.backend.qwen3.label" = "Qwen3-ASR 0.6B (CoreML)";
"asr.backend.qwen3.blurb" = "Stronger on Chinese dialects and noisy audio. ~1.6 GB CoreML model, works in background.";
"asr.backend.qwen3.downloadHint" = "Download the CoreML model in the section above before dictating.";
"asr.error.modelNotDownloaded" = "Qwen3-ASR is not downloaded yet. Download it from On-device models in Settings.";
"asr.error.unsupportedOS" = "Qwen3-ASR CoreML requires iOS 18 or later. Use iOS SpeechAnalyzer or switch to a cloud engine.";
/* App group error */
"appGroup.error.title" = "App Group not configured";
@@ -143,6 +197,22 @@
"preview.placeholder" = "Type here or tap the record button";
"preview.clear" = "Clear text";
"preview.openSettingsA11y" = "Open OSGKeyboard settings";
"preview.openSettings" = "Settings";
"preview.record.start" = "Start Recording";
"preview.record.stop" = "Stop Recording";
"preview.status.field" = "Text field";
"preview.status.recording" = "Recording…";
"preview.status.processing" = "Transcribing…";
/* Dictation handoff */
"dictation.title" = "Voice Input";
"dictation.status.ready" = "Ready to record…";
"dictation.status.listening" = "Listening…";
"dictation.status.listeningLive" = "Live transcription…";
"dictation.status.processing" = "Processing…";
"dictation.status.requestingPermission" = "Requesting permission…";
"dictation.status.waitingResult" = "Waiting for transcription…";
"dictation.status.saving" = "Saving…";
"preview.modeChip.cycle" = "Cycle input mode";
"preview.localeChip.cycle" = "Cycle recognition language";
"preview.error.audioSession" = "Audio session error: %@";
@@ -157,6 +227,7 @@
"keyboard.placeholder.processing" = "Processing";
"keyboard.placeholder.error" = "Polishing failed";
"keyboard.placeholder.localBadge" = "On-device";
"keyboard.placeholder.cloudBadge" = "Cloud";
"keyboard.rec" = "REC";
"keyboard.space" = "Space";
"keyboard.denied.mic" = "Mic denied";
@@ -206,6 +277,14 @@
"home.flow.endShort" = "End";
"home.preview.label" = "Try typing";
"home.preview.placeholder" = "Tap to type and test…";
"home.engine.unsupportedOS" = "Qwen3-ASR CoreML requires iOS 18+";
"home.engine.warming" = "Loading ASR model into memory…";
"home.engine.downloading" = "Downloading ASR model…";
"home.engine.loadFailed" = "Model downloaded but failed to load. Restart the app and try again.";
"home.engine.modelsReady" = "ASR model ready";
"home.engine.setupRequired" = "Local ASR model not ready";
"home.engine.downloadFirst" = "Download the local ASR model in Settings first";
"home.engine.attentionBanner" = "Local ASR model is not ready yet. Download it and wait for warm-up before dictating.";
/* Tabs */
"tab.keyboard" = "Keyboard";
@@ -224,3 +303,11 @@
"flow.error.micRequired" = "Microphone access is required for background voice sessions.";
"flow.error.micUnavailable" = "Microphone is unavailable on this device.";
"flow.error.audioUnavailable" = "Could not start background audio.";
"settings.models.error.metadata" = "Download cache was corrupted or incomplete. Try switching the download source and download again.";
"settings.models.error.offline" = "Could not reach the model server. Check your network or try a different download source.";
"settings.models.confirm.downloadTitle %@" = "Download %@";
"settings.models.confirm.downloadingTitleFormat %@" = "Downloading %@";
"settings.models.confirm.body %lld" = "This download uses about %lld MB on disk. We recommend Wi‑Fi to avoid cellular data. You can delete the file anytime from On-device models to free space.";
"settings.models.confirm.cancel" = "Cancel";
"settings.models.confirm.download %lld" = "Download %lld MB";