forked from Blendman974/faceclaw
Optimize APK size
Removes unused keyword-spotting model. Makes on-device voice model a runtime download rather than bundling it. Removes native-code library versions for 32-bit and x86_64 processors.
This commit is contained in:
1 parent
69ec828cb7
commit
2b71eddab5
27 files changed
+313
-1283
No files matched your search
@@ -195,7 +195,6 @@ class DashboardController {
|
||||
private offFrameMetrics: (() => void) | null = null;
|
||||
private offFirmwareInfo: (() => void) | null = null;
|
||||
private offVoiceStatus: (() => void) | null = null;
|
||||
private offVoiceWakeWord: (() => void) | null = null;
|
||||
private offAndroidNotification: (() => void) | null = null;
|
||||
private lastInput = "waiting...";
|
||||
private lastSys = "none yet";
|
||||
@@ -1011,11 +1010,6 @@ class DashboardController {
|
||||
this.offVoiceStatus = voiceControlBridge.onStatus((state) => {
|
||||
this.appendLog(state.status);
|
||||
});
|
||||
this.offVoiceWakeWord = voiceControlBridge.onWakeWord((keyword) => {
|
||||
void this.handleWakeWord(keyword).catch((error) => {
|
||||
this.appendLog(`wake-word handler failed: ${this.formatError(error)}`);
|
||||
});
|
||||
});
|
||||
|
||||
await mediaControllerBridge.start();
|
||||
await nightscoutBridge.start();
|
||||
@@ -1082,8 +1076,6 @@ class DashboardController {
|
||||
this.offFirmwareInfo = null;
|
||||
this.offVoiceStatus?.();
|
||||
this.offVoiceStatus = null;
|
||||
this.offVoiceWakeWord?.();
|
||||
this.offVoiceWakeWord = null;
|
||||
await mediaControllerBridge.stop().catch(() => {});
|
||||
await nightscoutBridge.stop().catch(() => {});
|
||||
voiceControlBridge.stop();
|
||||
@@ -1182,8 +1174,6 @@ class DashboardController {
|
||||
this.offFirmwareInfo = null;
|
||||
this.offVoiceStatus?.();
|
||||
this.offVoiceStatus = null;
|
||||
this.offVoiceWakeWord?.();
|
||||
this.offVoiceWakeWord = null;
|
||||
|
||||
const communicator = this.communicator;
|
||||
this.communicator = null;
|
||||
@@ -1816,20 +1806,6 @@ class DashboardController {
|
||||
this.updateCompositePreview();
|
||||
}
|
||||
|
||||
private async handleWakeWord(keyword: string): Promise<void> {
|
||||
const normalized = keyword.trim();
|
||||
if (normalized.length === 0) return;
|
||||
this.appendLog(`wake-word detected: ${normalized}`);
|
||||
const woke = shell.wake("sidebar");
|
||||
if (woke) {
|
||||
const ready = await this.ensureEvenHubSessionActive();
|
||||
if (!ready) {
|
||||
this.appendLog("EvenHub wake barrier timed out for phone wakeword");
|
||||
}
|
||||
this.requestShellRender();
|
||||
}
|
||||
}
|
||||
|
||||
private async handleAndroidNotificationPosted(notificationKey: string): Promise<void> {
|
||||
if (!notificationKey) {
|
||||
this.requestShellRender();
|
||||
|
||||
@@ -0,0 +1,176 @@
|
||||
import { Utils } from "@nativescript/core";
|
||||
|
||||
declare const com: any;
|
||||
declare const java: any;
|
||||
|
||||
/**
|
||||
* Download management for the on-device transcription model (Moonshine, via
|
||||
* sherpa-onnx). The model is no longer bundled in the APK; it is fetched on
|
||||
* demand into the same filesDir location that earlier releases copied the
|
||||
* bundled files to, so upgraded installs that already used on-device
|
||||
* transcription need no download. FaceclawVoiceController reads the files
|
||||
* from this directory.
|
||||
*
|
||||
* Mirrors the on-phone assistant model flow in llama.ts, except the model is
|
||||
* three files rather than one; they download sequentially through
|
||||
* FaceclawModelDownloader (resume + pinned sha256 per file).
|
||||
*/
|
||||
|
||||
type AsrModelFile = {
|
||||
name: string;
|
||||
sha256: string;
|
||||
sizeBytes: number;
|
||||
};
|
||||
|
||||
export const ASR_MODEL = {
|
||||
label: "Moonshine (English)",
|
||||
dirName: "sherpa-onnx-moonshine-base-en-quantized-2026-02-27",
|
||||
// Hugging Face mirror of the sherpa-onnx release asset of the same name.
|
||||
// The GitHub release only offers a tar.bz2, which the phone can't unpack;
|
||||
// this mirror serves the same files (sha256-verified) individually.
|
||||
baseUrl:
|
||||
"https://huggingface.co/csukuangfj2/sherpa-onnx-moonshine-base-en-quantized-2026-02-27/resolve/main/",
|
||||
files: [
|
||||
{
|
||||
name: "decoder_model_merged.ort",
|
||||
sha256: "d9d7b333af34bc552580576ddcf248a1c6c839e0d3b43b09afb9376ed009899d",
|
||||
sizeBytes: 109424400,
|
||||
},
|
||||
{
|
||||
name: "encoder_model.ort",
|
||||
sha256: "7c66495948d0d08ec1af454cd4b5514862ae6511e94712a60e6d83eaec8dc8cf",
|
||||
sizeBytes: 31326816,
|
||||
},
|
||||
{
|
||||
name: "tokens.txt",
|
||||
sha256: "2870d843e14c1e187bf1913a521562a63b53933814bd7f2145120468f494a049",
|
||||
sizeBytes: 549350,
|
||||
},
|
||||
] as AsrModelFile[],
|
||||
totalBytes: 141300566,
|
||||
};
|
||||
|
||||
export type AsrModelState = {
|
||||
status: "absent" | "downloading" | "ready";
|
||||
bytesDownloaded: number;
|
||||
totalBytes: number;
|
||||
};
|
||||
|
||||
let downloader: any = null;
|
||||
// Bytes of files already fully downloaded in this run, plus progress within
|
||||
// the file currently downloading; drives the aggregate percentage.
|
||||
let completedBytes = 0;
|
||||
let currentFileBytes = 0;
|
||||
const stateListeners = new Set<(state: AsrModelState) => void>();
|
||||
|
||||
function modelDirPath(): string {
|
||||
const context = Utils.android.getApplicationContext();
|
||||
return `${context.getFilesDir().getAbsolutePath()}/faceclaw-voice-asr/${ASR_MODEL.dirName}`;
|
||||
}
|
||||
|
||||
function filePath(file: AsrModelFile): string {
|
||||
return `${modelDirPath()}/${file.name}`;
|
||||
}
|
||||
|
||||
function isFilePresent(file: AsrModelFile): boolean {
|
||||
try {
|
||||
const javaFile = new java.io.File(filePath(file));
|
||||
return javaFile.exists() && javaFile.length() > 0;
|
||||
} catch {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
export function isAsrModelReady(): boolean {
|
||||
if (!global.isAndroid) return false;
|
||||
return ASR_MODEL.files.every(isFilePresent);
|
||||
}
|
||||
|
||||
export function asrModelState(): AsrModelState {
|
||||
if (downloader) {
|
||||
return {
|
||||
status: "downloading",
|
||||
bytesDownloaded: completedBytes + currentFileBytes,
|
||||
totalBytes: ASR_MODEL.totalBytes,
|
||||
};
|
||||
}
|
||||
return {
|
||||
status: isAsrModelReady() ? "ready" : "absent",
|
||||
bytesDownloaded: 0,
|
||||
totalBytes: ASR_MODEL.totalBytes,
|
||||
};
|
||||
}
|
||||
|
||||
export function onAsrModelStateChanged(listener: (state: AsrModelState) => void): () => void {
|
||||
stateListeners.add(listener);
|
||||
return () => stateListeners.delete(listener);
|
||||
}
|
||||
|
||||
function notifyStateChanged(): void {
|
||||
const state = asrModelState();
|
||||
stateListeners.forEach((listener) => listener(state));
|
||||
}
|
||||
|
||||
export function startAsrModelDownload(): void {
|
||||
if (!global.isAndroid || downloader || isAsrModelReady()) return;
|
||||
completedBytes = ASR_MODEL.files.filter(isFilePresent).reduce((sum, f) => sum + f.sizeBytes, 0);
|
||||
downloadNextFile();
|
||||
notifyStateChanged();
|
||||
}
|
||||
|
||||
function downloadNextFile(): void {
|
||||
const nextFile = ASR_MODEL.files.find((file) => !isFilePresent(file));
|
||||
if (!nextFile) {
|
||||
downloader = null;
|
||||
notifyStateChanged();
|
||||
return;
|
||||
}
|
||||
currentFileBytes = 0;
|
||||
const listener = new com.faceclaw.app.FaceclawModelDownloaderListener({
|
||||
onProgress: (bytes: number, _total: number) => {
|
||||
currentFileBytes = Number(bytes);
|
||||
notifyStateChanged();
|
||||
},
|
||||
onDone: () => {
|
||||
completedBytes += nextFile.sizeBytes;
|
||||
currentFileBytes = 0;
|
||||
downloadNextFile();
|
||||
},
|
||||
onError: (message: string) => {
|
||||
console.error(`Voice model download failed (${nextFile.name}): ${message}`);
|
||||
downloader = null;
|
||||
notifyStateChanged();
|
||||
},
|
||||
});
|
||||
downloader = new com.faceclaw.app.FaceclawModelDownloader(
|
||||
`${ASR_MODEL.baseUrl}${nextFile.name}`,
|
||||
filePath(nextFile),
|
||||
nextFile.sha256,
|
||||
nextFile.sizeBytes,
|
||||
listener,
|
||||
);
|
||||
downloader.start();
|
||||
}
|
||||
|
||||
/** Stops the download; already-fetched bytes are kept and resumed next time. */
|
||||
export function cancelAsrModelDownload(): void {
|
||||
if (!downloader) return;
|
||||
downloader.cancel();
|
||||
downloader = null;
|
||||
notifyStateChanged();
|
||||
}
|
||||
|
||||
export function deleteAsrModel(): void {
|
||||
if (!global.isAndroid) return;
|
||||
cancelAsrModelDownload();
|
||||
try {
|
||||
for (const file of ASR_MODEL.files) {
|
||||
new java.io.File(filePath(file)).delete();
|
||||
new java.io.File(`${filePath(file)}.part`).delete();
|
||||
}
|
||||
new java.io.File(modelDirPath()).delete();
|
||||
} catch (error) {
|
||||
console.error(`Voice model delete failed: ${String(error)}`);
|
||||
}
|
||||
notifyStateChanged();
|
||||
}
|
||||
@@ -45,7 +45,6 @@ export type RawPcmListener = (pcm: Uint8Array) => void;
|
||||
|
||||
export class FaceclawVoiceControlBridge {
|
||||
private readonly statusListeners = new Set<(state: VoiceControlState) => void>();
|
||||
private readonly wakeWordListeners = new Set<(keyword: string) => void>();
|
||||
private readonly transcriptListeners = new Set<(event: VoiceTranscriptEvent) => void>();
|
||||
private readonly speechEndListeners = new Set<() => void>();
|
||||
private controller: any | null = null;
|
||||
@@ -72,11 +71,6 @@ export class FaceclawVoiceControlBridge {
|
||||
return () => this.statusListeners.delete(listener);
|
||||
}
|
||||
|
||||
onWakeWord(listener: (keyword: string) => void): () => void {
|
||||
this.wakeWordListeners.add(listener);
|
||||
return () => this.wakeWordListeners.delete(listener);
|
||||
}
|
||||
|
||||
onTranscript(listener: (event: VoiceTranscriptEvent) => void): () => void {
|
||||
this.transcriptListeners.add(listener);
|
||||
return () => this.transcriptListeners.delete(listener);
|
||||
@@ -266,11 +260,6 @@ export class FaceclawVoiceControlBridge {
|
||||
if (this.rawActive) return;
|
||||
this.setStatus(String(status));
|
||||
},
|
||||
onWakeWord: (keyword: string) => {
|
||||
for (const listener of this.wakeWordListeners) {
|
||||
listener(String(keyword));
|
||||
}
|
||||
},
|
||||
onTranscript: (text: string, isFinal: boolean) => {
|
||||
this.emitTranscript(String(text), Boolean(isFinal));
|
||||
},
|
||||
|
||||
@@ -319,7 +319,7 @@ export const voiceProviderSetting = new ConfigSettingEnum<VoiceProvider>({
|
||||
if (value === "soniox") return sonioxApiKeySetting.get().trim().length === 0;
|
||||
return false;
|
||||
},
|
||||
description: "Speech-to-text engine for voice input. ElevenLabs, Whisper, and Soniox are cloud services that need an API key, with significantly better accuracy than on-device transcription.",
|
||||
description: "Speech-to-text engine for voice input. ElevenLabs, Whisper, and Soniox are cloud services that need an API key, with significantly better accuracy than on-device transcription. On-device transcription needs the voice model downloaded (below).",
|
||||
});
|
||||
|
||||
const wakeWordActionLabels: Record<WakeWordAction, string> = {
|
||||
|
||||
@@ -12,6 +12,14 @@ import {
|
||||
onLocalModelStateChanged,
|
||||
startLocalModelDownload,
|
||||
} from "../../native/llama";
|
||||
import {
|
||||
ASR_MODEL,
|
||||
asrModelState,
|
||||
cancelAsrModelDownload,
|
||||
deleteAsrModel,
|
||||
onAsrModelStateChanged,
|
||||
startAsrModelDownload,
|
||||
} from "../../native/asr-model";
|
||||
import { TextViewerLayer } from "../../apps/files/text-viewer";
|
||||
import type { LayerContext } from "../layers";
|
||||
import { drawRightValueMenuItem, openModalMenu, type MenuItem } from "../menu";
|
||||
@@ -87,6 +95,7 @@ function settingsSections(): SettingsSection[] {
|
||||
items: [
|
||||
enumSettingMenuItem(wakeWordActionSetting),
|
||||
enumSettingMenuItem(voiceProviderSetting),
|
||||
asrModelMenuItem(),
|
||||
],
|
||||
},
|
||||
{
|
||||
@@ -240,6 +249,75 @@ function localModelMenuItem(): MenuItem {
|
||||
};
|
||||
}
|
||||
|
||||
const ASR_MODEL_MB = `${Math.round(ASR_MODEL.totalBytes / 1e6)}MB`;
|
||||
|
||||
let asrModelRenderUnsub: (() => void) | null = null;
|
||||
|
||||
function watchAsrModelDownload(ctx: LayerContext): void {
|
||||
asrModelRenderUnsub?.();
|
||||
asrModelRenderUnsub = onAsrModelStateChanged((state) => {
|
||||
ctx.actions.requestRender();
|
||||
if (state.status !== "downloading") {
|
||||
asrModelRenderUnsub?.();
|
||||
asrModelRenderUnsub = null;
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
function asrModelStatusText(): string {
|
||||
const state = asrModelState();
|
||||
if (state.status === "ready") return "downloaded";
|
||||
if (state.status === "downloading") {
|
||||
const pct = state.totalBytes > 0 ? Math.floor((state.bytesDownloaded / state.totalBytes) * 100) : 0;
|
||||
return `${pct}% of ${ASR_MODEL_MB}`;
|
||||
}
|
||||
return "not downloaded";
|
||||
}
|
||||
|
||||
/** Download/cancel/delete management for the on-device transcription model. */
|
||||
function asrModelMenuItem(): MenuItem {
|
||||
return {
|
||||
label: "On-device voice model",
|
||||
description:
|
||||
`${ASR_MODEL.label} (${ASR_MODEL_MB} download). ` +
|
||||
"Transcribes voice input on the phone itself, with no API key or cloud service. " +
|
||||
"Required for the On-device transcription provider; the cloud providers work without it. " +
|
||||
"An interrupted download resumes where it left off.",
|
||||
onSelect: (ctx) => {
|
||||
const state = asrModelState();
|
||||
const action: MenuItem =
|
||||
state.status === "downloading"
|
||||
? {
|
||||
label: "Cancel download",
|
||||
onSelect: (innerCtx) => {
|
||||
cancelAsrModelDownload();
|
||||
innerCtx.stack.pop();
|
||||
},
|
||||
}
|
||||
: state.status === "ready"
|
||||
? {
|
||||
label: "Delete model",
|
||||
onSelect: (innerCtx) => {
|
||||
deleteAsrModel();
|
||||
innerCtx.stack.pop();
|
||||
},
|
||||
}
|
||||
: {
|
||||
label: `Download (${ASR_MODEL_MB})`,
|
||||
onSelect: (innerCtx) => {
|
||||
startAsrModelDownload();
|
||||
watchAsrModelDownload(innerCtx);
|
||||
innerCtx.stack.pop();
|
||||
},
|
||||
};
|
||||
openModalMenu(ctx, "On-device voice model", [action], 0);
|
||||
},
|
||||
render: ({ image, x, y, width }) => {
|
||||
drawRightValueMenuItem(image, getDefaultSmallFont(), x, y, width, "On-device voice model", asrModelStatusText());
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
/** A row that opens one of the project docs (copied into the bundle under
|
||||
* about/ by webpack.config.js) in the paged text viewer. */
|
||||
function bundledDocMenuItem(fileName: string, label: string): MenuItem {
|
||||
|
||||
Reference in new issue
Block a user