mirror of
https://github.com/legop3/MultiRoombaRover.git
synced 2026-09-16 01:21:20 -04:00
adding barcodesses. piper slop tts
This commit is contained in:
legop3
@@ -1,114 +0,0 @@
|
||||
// Kokoro Browser TTS Adapter
|
||||
// Purpose: Runs scanner speech through Kokoro.js locally in the browser when available.
|
||||
// Scope: Keeps the heavy model dependency behind a dynamic import so non-scanner pages do not load it.
|
||||
|
||||
const KOKORO_MODEL_ID = 'onnx-community/Kokoro-82M-v1.0-ONNX';
|
||||
const KOKORO_OPTIONS = {
|
||||
// q8 is the practical first pass for a scanner station: much smaller than
|
||||
// fp32/fp16, still good enough for short object labels, and it runs on plain
|
||||
// WASM so Firefox does not need WebGPU support.
|
||||
dtype: 'q8',
|
||||
device: 'wasm',
|
||||
};
|
||||
const KOKORO_VOICE = 'af_bella';
|
||||
const KOKORO_SPEED = 1;
|
||||
|
||||
let kokoroModelPromise = null;
|
||||
let currentAudio = null;
|
||||
let currentObjectUrl = null;
|
||||
|
||||
function stopCurrentAudio() {
|
||||
if (currentAudio) {
|
||||
// Scanner scans can arrive close together. Stop the previous generated clip
|
||||
// before starting the next one so the spoken label always matches the most
|
||||
// recent server scan state.
|
||||
currentAudio.pause();
|
||||
currentAudio.removeAttribute('src');
|
||||
currentAudio.load();
|
||||
currentAudio = null;
|
||||
}
|
||||
if (currentObjectUrl) {
|
||||
URL.revokeObjectURL(currentObjectUrl);
|
||||
currentObjectUrl = null;
|
||||
}
|
||||
}
|
||||
|
||||
async function loadKokoroModel() {
|
||||
if (!kokoroModelPromise) {
|
||||
kokoroModelPromise = import('kokoro-js')
|
||||
.then(async ({ KokoroTTS }) => {
|
||||
console.info('[scanner tts] loading Kokoro model');
|
||||
const model = await KokoroTTS.from_pretrained(KOKORO_MODEL_ID, {
|
||||
...KOKORO_OPTIONS,
|
||||
progress_callback: (progress) => {
|
||||
// The normal scanner page stays clean, but console progress is very
|
||||
// useful when first-load model downloads are slow on the scanner PC.
|
||||
if (progress?.status) {
|
||||
console.info('[scanner tts] Kokoro load', progress);
|
||||
}
|
||||
},
|
||||
});
|
||||
console.info('[scanner tts] Kokoro model ready');
|
||||
return model;
|
||||
})
|
||||
.catch((err) => {
|
||||
// Reset the promise after a failure so a later scan can retry. This is
|
||||
// important for first-load network hiccups while still letting the hook
|
||||
// fall back to browser speech immediately.
|
||||
kokoroModelPromise = null;
|
||||
throw err;
|
||||
});
|
||||
}
|
||||
return kokoroModelPromise;
|
||||
}
|
||||
|
||||
function playBlob(blob) {
|
||||
stopCurrentAudio();
|
||||
currentObjectUrl = URL.createObjectURL(blob);
|
||||
currentAudio = new Audio(currentObjectUrl);
|
||||
currentAudio.volume = 1;
|
||||
|
||||
return new Promise((resolve, reject) => {
|
||||
const audio = currentAudio;
|
||||
audio.onended = () => {
|
||||
if (currentAudio === audio) {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
resolve(true);
|
||||
};
|
||||
audio.onerror = () => {
|
||||
if (currentAudio === audio) {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
reject(new Error('Kokoro audio playback failed'));
|
||||
};
|
||||
audio.play().catch((err) => {
|
||||
if (currentAudio === audio) {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
reject(err);
|
||||
});
|
||||
});
|
||||
}
|
||||
|
||||
export async function speakWithKokoro(text) {
|
||||
const cleanText = String(text || '').replace(/\s+/g, ' ').trim();
|
||||
if (!cleanText) return false;
|
||||
|
||||
try {
|
||||
const model = await loadKokoroModel();
|
||||
const audio = await model.generate(cleanText, {
|
||||
voice: KOKORO_VOICE,
|
||||
speed: KOKORO_SPEED,
|
||||
});
|
||||
await playBlob(audio.toBlob());
|
||||
return true;
|
||||
} catch (err) {
|
||||
console.warn('[scanner tts] Kokoro unavailable; falling back to browser speech', err);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
export function stopKokoroSpeech() {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
@@ -0,0 +1,117 @@
|
||||
// Piper Browser TTS Adapter
|
||||
// Purpose: Runs scanner speech through Piper locally in the browser when available.
|
||||
// Scope: Keeps the neural TTS package behind a dynamic import so non-scanner pages do not load it.
|
||||
|
||||
const PIPER_VOICE_ID = 'en_US-hfc_female-medium';
|
||||
const PIPER_LIVE_TIMEOUT_MS = 900;
|
||||
|
||||
let piperModulePromise = null;
|
||||
let currentAudio = null;
|
||||
let currentObjectUrl = null;
|
||||
let activeRequestId = 0;
|
||||
|
||||
function stopCurrentAudio() {
|
||||
if (currentAudio) {
|
||||
// Scanner labels should never overlap. Stop the previous clip before the
|
||||
// next scan speaks so audio always matches the current large display text.
|
||||
currentAudio.pause();
|
||||
currentAudio.removeAttribute('src');
|
||||
currentAudio.load();
|
||||
currentAudio = null;
|
||||
}
|
||||
if (currentObjectUrl) {
|
||||
URL.revokeObjectURL(currentObjectUrl);
|
||||
currentObjectUrl = null;
|
||||
}
|
||||
}
|
||||
|
||||
async function loadPiperModule() {
|
||||
if (!piperModulePromise) {
|
||||
piperModulePromise = import('@mintplex-labs/piper-tts-web')
|
||||
.then((module) => {
|
||||
console.info('[scanner tts] Piper module ready');
|
||||
return module;
|
||||
})
|
||||
.catch((err) => {
|
||||
// Reset after failures so later scans can try again after a transient
|
||||
// model/CDN/cache issue. The hook falls back to browser speech for the
|
||||
// current scan rather than blocking the scanner interaction.
|
||||
piperModulePromise = null;
|
||||
throw err;
|
||||
});
|
||||
}
|
||||
return piperModulePromise;
|
||||
}
|
||||
|
||||
function playBlob(blob) {
|
||||
stopCurrentAudio();
|
||||
currentObjectUrl = URL.createObjectURL(blob);
|
||||
currentAudio = new Audio(currentObjectUrl);
|
||||
currentAudio.volume = 1;
|
||||
|
||||
return new Promise((resolve, reject) => {
|
||||
const audio = currentAudio;
|
||||
audio.onended = () => {
|
||||
if (currentAudio === audio) {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
resolve(true);
|
||||
};
|
||||
audio.onerror = () => {
|
||||
if (currentAudio === audio) {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
reject(new Error('Piper audio playback failed'));
|
||||
};
|
||||
audio.play().catch((err) => {
|
||||
if (currentAudio === audio) {
|
||||
stopCurrentAudio();
|
||||
}
|
||||
reject(err);
|
||||
});
|
||||
});
|
||||
}
|
||||
|
||||
export async function speakWithPiper(text) {
|
||||
const cleanText = String(text || '').replace(/\s+/g, ' ').trim();
|
||||
if (!cleanText) return false;
|
||||
const requestId = activeRequestId + 1;
|
||||
activeRequestId = requestId;
|
||||
|
||||
try {
|
||||
const wavBlob = await Promise.race([
|
||||
loadPiperModule().then((piper) =>
|
||||
piper.predict(
|
||||
{
|
||||
text: cleanText,
|
||||
voiceId: PIPER_VOICE_ID,
|
||||
},
|
||||
(progress) => {
|
||||
// Piper stores downloaded models in the browser origin-private file
|
||||
// system. Console progress is enough for setup/debugging while the
|
||||
// rover-facing scanner screen stays uncluttered.
|
||||
if (progress?.url) {
|
||||
console.info('[scanner tts] Piper load', progress);
|
||||
}
|
||||
},
|
||||
),
|
||||
),
|
||||
new Promise((resolve) => {
|
||||
window.setTimeout(() => resolve(null), PIPER_LIVE_TIMEOUT_MS);
|
||||
}),
|
||||
]);
|
||||
if (!wavBlob || requestId !== activeRequestId) {
|
||||
return false;
|
||||
}
|
||||
await playBlob(wavBlob);
|
||||
return true;
|
||||
} catch (err) {
|
||||
console.warn('[scanner tts] Piper unavailable; falling back to browser speech', err);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
export function stopPiperSpeech() {
|
||||
activeRequestId += 1;
|
||||
stopCurrentAudio();
|
||||
}
|
||||
@@ -2,7 +2,7 @@
|
||||
// Purpose: Speaks server-provided scan labels on the scanner computer.
|
||||
// Scope: Keeps browser TTS as a local output device while leaving scan meaning and phrase selection on the server.
|
||||
import { useEffect, useRef } from 'react';
|
||||
import { speakWithKokoro, stopKokoroSpeech } from './scannerTts/kokoroBrowserTts.js';
|
||||
import { speakWithPiper, stopPiperSpeech } from './scannerTts/piperBrowserTts.js';
|
||||
|
||||
const SPEECH_START_TIMEOUT_MS = 1500;
|
||||
const SPEECH_RETRY_DELAY_MS = 700;
|
||||
@@ -54,7 +54,7 @@ export default function useScannerSpeech(scan) {
|
||||
return () => {
|
||||
window.clearTimeout(retryTimerRef.current);
|
||||
window.clearTimeout(startTimerRef.current);
|
||||
stopKokoroSpeech();
|
||||
stopPiperSpeech();
|
||||
};
|
||||
}, []);
|
||||
|
||||
@@ -123,11 +123,12 @@ export default function useScannerSpeech(scan) {
|
||||
function speakOnce() {
|
||||
if (cancelled) return;
|
||||
clearSpeechTimers();
|
||||
stopKokoroSpeech();
|
||||
// Kokoro is attempted first because it gives the scanner page a consistent
|
||||
// local model voice instead of relying on the OS/browser Web Speech voice
|
||||
// list. The old browser TTS path remains the reliability fallback.
|
||||
speakWithKokoro(speechText).then((spoken) => {
|
||||
stopPiperSpeech();
|
||||
// Piper is attempted first because it is lighter than Kokoro while still
|
||||
// giving the scanner page a consistent local model voice. The old browser
|
||||
// TTS path remains the reliability fallback if Piper is still loading or
|
||||
// fails on the scanner computer.
|
||||
speakWithPiper(speechText).then((spoken) => {
|
||||
if (cancelled || spoken) return;
|
||||
speakWithBrowserFallback();
|
||||
});
|
||||
@@ -138,7 +139,7 @@ export default function useScannerSpeech(scan) {
|
||||
return () => {
|
||||
cancelled = true;
|
||||
clearSpeechTimers();
|
||||
stopKokoroSpeech();
|
||||
stopPiperSpeech();
|
||||
};
|
||||
}, [scan]);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user