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 { useEffect, useRef, useState } from 'react';
import {
UploadCloud,
X,
ArrowRightLeft,
Loader,
AudioLines,
Fingerprint,
Timer,
Info,
} from 'lucide-react';
import { toast } from 'react-hot-toast';
import { Button } from '../../ui';
import { API } from '../../api/client';
import { convertSpeech } from '../../api/convert';
import { TtsGenerationBusyError } from '../../api/generate';
import { asrMissingPayload, toastAsrModelMissing } from '../../utils/asrModelMissing';
import { modelNotDownloadedPayload, toastModelNotDownloaded } from '../../utils/modelNotDownloaded';
import { toastErrorWithReport } from '../../utils/errorToast';
import useRecording from '../../hooks/useRecording';
import MicButton from './MicButton';
import VoiceSelector from '../VoiceSelector';
import WaveformPlayer from '../WaveformPlayer';
/**
* Studio "Convert" method — speech-to-speech voice changer.
*
* Drop or record a source clip, pick an existing voice profile, Convert:
* the backend transcribes the clip with the active ASR engine and re-says
* it with the active TTS engine in the chosen profile's voice
* (POST /convert). "Match source duration" (default on) time-stretches the
* take — pitch preserved — to land near the source clip's length.
*
* Self-contained on purpose: unlike the audio/design methods it doesn't use
* ScriptPanel (the script IS the source clip) or the shared ActionBar, so it
* owns its source file, target voice, and result state locally.
*/
export default function ConvertMethodPanel({ t, profiles = [], onRecordingBusyChange }) {
const [sourceFile, setSourceFile] = useState(null);
const [voiceId, setVoiceId] = useState('');
const [matchDuration, setMatchDuration] = useState(true);
const [isConverting, setIsConverting] = useState(false);
const [result, setResult] = useState(null);
// Stale-request guard: every source/voice change bumps the sequence AND
// aborts any in-flight convert, so an obsolete take never renders against
// the newly-chosen inputs and the abort releases the shared TTS admission
// slot — the next convert isn't blocked behind a request nobody wants.
const convertSeqRef = useRef(0);
const abortRef = useRef(null);
const invalidateInFlight = () => {
convertSeqRef.current += 1;
const obsolete = abortRef.current;
if (obsolete) {
abortRef.current = null;
obsolete.abort();
setIsConverting(false);
}
};
const ingestSource = (file) => {
if (!file) return;
if (
!(file.type.startsWith('audio/') || /\.(mp3|wav|m4a|flac|ogg|aac|webm)$/i.test(file.name))
) {
toast.error(t('clone.unsupported_audio'));
return;
}
invalidateInFlight();
// Re-wrap the picked/dropped File with a metacharacter-free name before it
// enters state (CodeQL js/xss-through-dom): a file's NAME is DOM-derived
// text, and this object later feeds the shared player and the upload
// filename. Same bytes, same type — only the name is constrained.
const safeName = (file.name || 'source.wav').replace(/[^\w .()-]+/g, '_');
setSourceFile(new File([file], safeName, { type: file.type }));
setResult(null);
};
const selectVoice = (id) => {
invalidateInFlight();
setVoiceId(id);
// A finished take belongs to the previously chosen voice — showing it
// against the new selection would misattribute the audio.
setResult(null);
};
// Own mic instance: a convert source is not the clone reference, so it
// must never overwrite the audio method's refAudio.
const {
isRecording,
isStartingRecording,
isCleaning,
recordingTime,
startRecording,
stopRecording,
} = useRecording(async (file) => ingestSource(file));
useEffect(() => {
onRecordingBusyChange?.(
Boolean(isStartingRecording || isRecording || isCleaning || isConverting),
);
return () => onRecordingBusyChange?.(false);
}, [isStartingRecording, isRecording, isCleaning, isConverting, onRecordingBusyChange]);
const canConvert = !!sourceFile && !!voiceId && !isConverting;
const handleConvert = async () => {
if (!canConvert) return;
const seq = convertSeqRef.current;
const controller = new AbortController();
abortRef.current = controller;
setIsConverting(true);
setResult(null);
try {
const fd = new FormData();
// Re-wrap the File as a fresh Blob — same Tauri/WebKit quirk workaround
// as useTTS's ref_audio append (a stale File handle can upload empty).
const buf = await sourceFile.arrayBuffer();
// File reads are not abortable. If the inputs changed while this read
// was pending, stop before the obsolete source can enter admission.
if (seq !== convertSeqRef.current || controller.signal.aborted) return;
fd.append(
'audio',
new Blob([buf], { type: sourceFile.type || 'audio/wav' }),
sourceFile.name || 'source.wav',
);
fd.append('profile_id', voiceId);
fd.append('match_duration', matchDuration ? '1' : '0');
const res = await convertSpeech(fd, { signal: controller.signal });
if (seq === convertSeqRef.current) setResult(res);
} catch (e) {
// Obsolete request (inputs changed, which also aborted it) — silent.
if (seq !== convertSeqRef.current || e?.name === 'AbortError') return;
// Same error ladder as useTTS's generate catch: structured payloads get
// their one-click CTA, the busy guard gets its localized notice, and
// everything else goes through toastErrorWithReport — which maps the
// backend's [shutting_down]/[clone_ref_*] markers to localized guidance
// and gives real failures the "Report" action instead of a raw string.
const missing = asrMissingPayload(e);
const notDownloaded = modelNotDownloadedPayload(e);
if (missing) {
toastAsrModelMissing(missing);
} else if (notDownloaded) {
toastModelNotDownloaded(notDownloaded);
} else if (e instanceof TtsGenerationBusyError) {
toast(t('tts_errors.generation_in_progress'), { icon: '⏳' });
} else {
toastErrorWithReport(t('tts_errors.error_prefix', { message: e?.message || String(e) }), e);
}
} finally {
// An obsolete request may finish after its replacement has started.
// Only the request that still owns the ref may release the busy state.
if (abortRef.current === controller) {
abortRef.current = null;
setIsConverting(false);
}
}
};
return (
<div data-testid="convert-method-panel" className="flex flex-1 min-h-0 flex-col">
<div
data-testid="convert-form"
className="convert-form flex-1 min-h-0 overflow-y-auto px-3 py-2 space-y-5"
>
{/* ── Source clip: drop / pick / record ── */}
<div className="flex items-center gap-2 text-sm font-medium text-[var(--chrome-fg)]">
<AudioLines size={17} aria-hidden="true" />
{t('convert.source_kicker')}
</div>
<div className="flex flex-wrap gap-3 items-stretch rounded-lg bg-[var(--chrome-hover-bg)] p-4">
<input
type="file"
accept="audio/*,.mp3,.wav,.m4a,.flac,.ogg"
onChange={(e) => {
ingestSource(e.target.files[0]);
e.target.value = '';
}}
className="sr-only"
id="convert-audio-upload"
/>
<label
htmlFor="convert-audio-upload"
tabIndex={0}
role="button"
onKeyDown={(event) => {
if (event.key === 'Enter' || event.key === ' ') {
event.preventDefault();
document.getElementById('convert-audio-upload')?.click();
}
}}
className="flex-1 basis-56 min-w-0 min-h-48 rounded-lg p-5 text-center cursor-pointer flex flex-col justify-center items-center gap-3 bg-[var(--chrome-hover-bg)] transition-colors hover:bg-[var(--chrome-accent-bg)] focus-visible:outline-2 focus-visible:outline-[var(--chrome-accent)] [&.is-dragging]:bg-[var(--chrome-accent-bg)]"
onDragOver={(e) => {
e.preventDefault();
e.currentTarget.classList.add('is-dragging');
}}
onDragLeave={(e) => {
e.currentTarget.classList.remove('is-dragging');
}}
onDrop={(e) => {
e.preventDefault();
e.currentTarget.classList.remove('is-dragging');
const file = e.dataTransfer.files[0];
ingestSource(file);
}}
>
<UploadCloud className="text-[var(--chrome-accent)]" size={28} aria-hidden="true" />
<p className="m-0 text-sm text-[color:var(--chrome-fg-muted)] font-[family-name:var(--font-sans)] font-medium">
{sourceFile ? (
<span className="text-fg">{sourceFile.name}</span>
) : (
t('convert.drop_audio')
)}
</p>
</label>
<MicButton
isCleaning={isCleaning}
isStarting={isStartingRecording}
isRecording={isRecording}
recordingTime={recordingTime}
onStart={startRecording}
onStop={stopRecording}
/>
</div>
{sourceFile && (
<div className="mt-2 flex items-center gap-[8px]">
<div className="flex-1 min-w-0">
<WaveformPlayer src={sourceFile} source="convert-source" height={34} compact />
</div>
<Button
variant="ghost"
size="sm"
onClick={() => {
invalidateInFlight();
setSourceFile(null);
setResult(null);
}}
leading={<X size={11} />}
>
{t('clone.clear')}
</Button>
</div>
)}
{/* ── Target voice ── */}
<div className="flex items-center gap-2 text-sm font-medium text-[var(--chrome-fg)]">
<Fingerprint size={17} aria-hidden="true" />
{t('convert.target_voice')}
</div>
<div className="rounded-lg bg-[var(--chrome-hover-bg)] p-3">
<VoiceSelector
value={voiceId}
onChange={selectVoice}
profiles={profiles}
engineDefault={false}
gallery={false}
placeholder={t('convert.pick_voice')}
ariaLabel={t('convert.target_voice')}
recentsKey="convert-target"
menuPortal
buttonClassName="min-h-12 px-3 text-sm border-0 rounded-lg bg-transparent text-[var(--chrome-fg)] hover:bg-[var(--chrome-accent-bg)] focus-visible:outline-2 focus-visible:outline-[var(--chrome-accent)]"
/>
</div>
{/* ── Options + action ── */}
<div className="flex flex-col items-start gap-3 rounded-lg bg-[var(--chrome-hover-bg)] p-4">
<label
className="inline-flex items-center gap-[6px] text-[0.85em] text-fg-muted cursor-pointer select-none whitespace-nowrap"
title={t('convert.match_duration_hint')}
>
<input
type="checkbox"
checked={matchDuration}
onChange={(e) => {
invalidateInFlight();
setMatchDuration(e.target.checked);
setResult(null);
}}
/>
<Timer size={16} aria-hidden="true" />
<span>{t('convert.match_duration')}</span>
</label>
<p className="m-0 text-xs text-[var(--chrome-fg-muted)]">
{t('convert.match_duration_hint')}
</p>
{!sourceFile || !voiceId ? (
<span className="inline-flex items-center gap-2 text-xs text-fg-muted">
<Info size={14} aria-hidden="true" />
{t('convert.need_source_and_voice')}
</span>
) : null}
</div>
{/* ── Result: converted take + what the ASR heard ── */}
{result && (
<div className="mt-[var(--space-4)]" data-testid="convert-result">
<div className="label-row">{t('convert.result_kicker')}</div>
<WaveformPlayer
src={`${API}${result.audio_url}`}
source="output"
height={40}
autoPlay
/>
<div className="mt-2 text-[0.78rem] text-fg-muted">
<span className="font-medium">{t('convert.transcript')}:</span> {result.text}
</div>
</div>
)}
</div>
<div className="studio-action-bar" data-testid="convert-action-bar">
<Button
variant="primary"
block
onClick={handleConvert}
disabled={!canConvert}
leading={
isConverting ? (
<Loader size={14} className="animate-spin" />
) : (
<ArrowRightLeft size={14} />
)
}
>
{isConverting ? t('convert.converting') : t('convert.convert')}
</Button>
</div>
</div>
);
}