VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.
Python aiaudiobookcudadubbingelevenlabs-alternativehuggingfacelocal-firstmlxomnivoice-studiospeech-to-texttext-to-speechtranscriptiontranslatettsvoice-aivoice-cloningvoice-generationvoicestudioworkflow
import { useCallback, useEffect, useMemo, useState } from 'react';
import { ClipboardPaste, FileText, AlertTriangle } from 'lucide-react';
import { useTranslation } from 'react-i18next';
import { Dialog, Button, Textarea, Badge } from '../../ui';
import { buildPastePlan, detectPasteMode } from '../../utils/pasteTranslations';
import { dubParseSubtitleText } from '../../api/dub';
import { readTextFile } from '../../utils/readTextFile';
/**
* DubPasteTranslationDialog — paste a translation produced somewhere else
* (ChatGPT, DeepL, a human translator) onto the segments that already exist.
*
* Nothing is applied until the user has seen the full before→after mapping:
* a wrong-by-one paste over a 400-line transcript is expensive to notice
* later and annoying to undo row by row. Unmatched rows are shown as such
* rather than being quietly filled with the neighbouring line.
*
* `onApply(text, { mode, cues })` re-derives the identical plan through
* `pasteTranslations`, so applying goes through the normal undo stack.
*/
const MODE_ICON = { timestamped: FileText, numbered: ClipboardPaste, plain: ClipboardPaste };
// A feature-length transcript runs to thousands of rows; mounting all of them
// in a 280px scroll box just to preview costs more than it informs. Show a
// window and say how many are hidden — the count banner above already carries
// the totals the decision actually rests on.
const PREVIEW_ROW_LIMIT = 200;
export default function DubPasteTranslationDialog({ open, segments = [], onApply, onClose }) {
const { t } = useTranslation();
const [text, setText] = useState('');
const [cues, setCues] = useState(null);
const [parsing, setParsing] = useState(false);
const [parseError, setParseError] = useState(null);
useEffect(() => {
if (open) {
setText('');
setCues(null);
setParseError(null);
}
}, [open]);
const mode = useMemo(() => detectPasteMode(text), [text]);
// Timestamped pastes need the backend's lenient cue parser before a plan
// can exist. Debounced so a long paste isn't re-parsed on every keystroke.
useEffect(() => {
if (!open || mode !== 'timestamped' || !text.trim()) {
setCues(null);
setParseError(null);
setParsing(false);
return undefined;
}
let cancelled = false;
// Drop the previous parse the moment the text changes: keeping it would
// let the preview — and an eager Apply during the debounce window — run
// against cues from text the user has already edited away.
setCues(null);
setParseError(null);
setParsing(true);
const timer = setTimeout(() => {
dubParseSubtitleText(text)
.then((res) => {
if (cancelled) return;
setCues(res.segments || []);
setParseError(null);
})
.catch((e) => {
if (cancelled) return;
setCues([]);
setParseError(e?.message || String(e));
})
.finally(() => {
if (!cancelled) setParsing(false);
});
}, 300);
return () => {
cancelled = true;
clearTimeout(timer);
};
}, [open, mode, text]);
const plan = useMemo(() => {
if (!text.trim()) return null;
if (mode === 'timestamped' && cues === null) return null;
return buildPastePlan(text, segments, { mode, cues: cues || [] });
}, [text, segments, mode, cues]);
const readFile = useCallback((file) => {
if (!file) return;
// As before, a file that cannot be read leaves the text as it was.
readTextFile(file).then(setText, () => {});
}, []);
const apply = () => {
onApply?.(text, { mode, cues: cues || [] });
onClose?.();
};
if (!open) return null;
const ModeIcon = MODE_ICON[mode] || ClipboardPaste;
const canApply = !!plan && plan.matchedCount > 0;
return (
<Dialog
open
onClose={onClose}
size="lg"
title={
<>
<ClipboardPaste size={14} /> {t('dub.paste_translation_title')}
</>
}
footer={
<>
<Button variant="ghost" onClick={onClose}>
{t('dub.paste_translation_cancel')}
</Button>
<Button variant="primary" onClick={apply} disabled={!canApply} loading={parsing}>
{t('dub.paste_translation_apply', { count: plan?.matchedCount || 0 })}
</Button>
</>
}
>
<p className="text-[length:var(--text-xs)] text-fg-muted mb-[var(--space-3)]">
{t('dub.paste_translation_desc')}
</p>
<Textarea
rows={7}
value={text}
onChange={(e) => setText(e.target.value)}
placeholder={t('dub.paste_translation_placeholder')}
aria-label={t('dub.paste_translation_title')}
autoFocus
onDrop={(e) => {
const file = e.dataTransfer?.files?.[0];
if (file) {
e.preventDefault();
readFile(file);
}
}}
/>
<div className="flex items-center gap-[var(--space-3)] mt-[var(--space-3)] flex-wrap">
<label
htmlFor="paste-translation-file"
className="flex items-center gap-[6px] px-[10px] py-[4px] rounded-[6px] cursor-pointer text-[length:var(--text-xs)] text-fg-muted bg-[rgba(255,255,255,0.05)] [border:1px_solid_rgba(255,255,255,0.1)]"
>
<FileText size={11} /> {t('dub.paste_translation_load_file')}
{/* `sr-only`, never `hidden`: a bare <label> isn't in the tab order,
and `hidden` drops the input from both the tab order and the
accessibility tree — together they leave keyboard-only users with
no way to reach the file picker at all. Visually hidden but
focusable keeps the pointer affordance and restores the keyboard
path (focus the input, Enter/Space opens the picker). */}
<input
id="paste-translation-file"
type="file"
accept=".srt,.vtt,.txt,text/plain"
className="sr-only"
onChange={(e) => {
readFile(e.target.files?.[0]);
e.target.value = '';
}}
/>
</label>
{!!text.trim() && (
<Badge tone="neutral" size="xs">
<ModeIcon size={10} /> {t(`dub.paste_translation_mode_${mode}`)}
</Badge>
)}
</div>
{parseError && (
<div
role="alert"
data-testid="paste-translation-error"
className="flex items-center gap-[6px] mt-[var(--space-3)] text-[length:var(--text-xs)] text-[#fb4934]"
>
<AlertTriangle size={11} /> {parseError}
</div>
)}
{plan && (
<>
<div
className="mt-[var(--space-4)] px-[var(--space-3)] py-[6px] rounded-[var(--radius-md)] text-[length:var(--text-xs)] bg-[rgba(255,255,255,0.03)] [border:1px_solid_rgba(255,255,255,0.06)]"
data-testid="paste-translation-summary"
>
{t('dub.paste_translation_counts', {
segments: plan.rows.length,
lines: plan.sourceCount,
unmatched: plan.unmatchedCount,
})}
{plan.unusedCount > 0 && (
<span className="text-[#fabd2f] ml-[var(--space-2)]">
{t('dub.paste_translation_unused', { count: plan.unusedCount })}
</span>
)}
</div>
{plan.matchedCount === 0 && (
<div
role="alert"
className="flex items-center gap-[6px] mt-[var(--space-2)] text-[length:var(--text-xs)] text-[#fb4934]"
>
<AlertTriangle size={11} /> {t('dub.paste_translation_none_matched')}
</div>
)}
<div className="mt-[var(--space-3)] max-h-[280px] overflow-y-auto [border:1px_solid_var(--chrome-border)] rounded-[var(--radius-md)]">
{plan.rows.slice(0, PREVIEW_ROW_LIMIT).map((row) => (
<div
key={row.id}
data-testid="paste-translation-row"
data-matched={row.matched ? 'true' : 'false'}
className={`flex gap-[var(--space-3)] px-[var(--space-3)] py-[4px] text-[length:var(--text-xs)] [border-bottom:1px_solid_var(--chrome-border)] ${
row.matched ? '' : 'bg-[rgba(251,73,52,0.06)]'
}`}
>
<span className="w-[28px] shrink-0 text-fg-muted font-[family-name:var(--chrome-font-mono)]">
{row.index + 1}
</span>
<span className="flex-1 min-w-0 text-fg-muted line-through decoration-[rgba(255,255,255,0.25)]">
{row.before || '—'}
</span>
<span className="flex-1 min-w-0 text-fg">
{row.matched ? (
row.after
) : (
<em className="text-[#fb4934] not-italic">
{t('dub.paste_translation_row_unmatched')}
</em>
)}
</span>
</div>
))}
{plan.rows.length > PREVIEW_ROW_LIMIT && (
<div className="px-[var(--space-3)] py-[6px] text-[length:var(--text-xs)] text-fg-muted text-center">
{t('dub.paste_translation_more_rows', {
count: plan.rows.length - PREVIEW_ROW_LIMIT,
})}
</div>
)}
</div>
</>
)}
</Dialog>
);
}