Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: send audio clips, stems and transcripts to other Audio pages (#12811)

Etherlfed78103abef
diff --git a/studio/frontend/src/features/audio/audio-page.tsx b/studio/frontend/src/features/audio/audio-page.tsx
index 07da40a5f37..ad217a64e73 100644
--- a/studio/frontend/src/features/audio/audio-page.tsx
+++ b/studio/frontend/src/features/audio/audio-page.tsx
@@ -55,13 +55,15 @@ import {
isMusicGenerationModel,
} from "./catalog";
import type { ClipSendHandlers } from "./components/clip-card";
+import { sendClipToMusic } from "./components/music-send-to";
+import { SaveVoiceDialog } from "./components/save-voice-dialog";
import { WorkflowTitleMenu } from "./components/workflow-title-menu";
import { galleryCache, useAudioGallery, useWorkflowHistory } from "./hooks/use-audio-gallery";
import { useAudioHandoff } from "./hooks/use-audio-handoff";
import { useAudioModelSlot } from "./hooks/use-audio-model-slot";
import { useCloneGeneration } from "./hooks/use-clone-generation";
import { useConvertGeneration } from "./hooks/use-convert-generation";
-import { useEditGeneration } from "./hooks/use-edit-generation";
+import { adoptEditSource, useEditGeneration } from "./hooks/use-edit-generation";
import { useSeparateGeneration } from "./hooks/use-separate-generation";
import { useSpeechGeneration } from "./hooks/use-speech-generation";
import { useSttSidecar } from "./hooks/use-stt-sidecar";
@@ -109,11 +111,18 @@ import {
transcribeLanguagesFor,
} from "./transcribe-languages";
import { type AudioPickerRow, audioRowMatchesWorkflow } from "./picker-filter";
-import { useAudioCloneStore } from "./stores/audio-clone-store";
+import {
+ referenceTranscript,
+ useAudioCloneStore,
+} from "./stores/audio-clone-store";
import { useAudioConvertStore } from "./stores/audio-convert-store";
import { useAudioEditStore } from "./stores/audio-edit-store";
+import { useAudioSeparateStore } from "./stores/audio-separate-store";
+import { useAudioVoicesStore } from "./stores/audio-voices-store";
import { useAudioWorkspaceStore } from "./stores/audio-workspace-store";
+import { clipSendTargets, transcriptSendTargets } from "./send-targets";
import { AudioToolPanels } from "./tools/tool-panel-host";
+import type { TranscriptSource } from "./transcript-model";
import {
type AudioWorkflowId,
audioWorkflowTab,
@@ -751,46 +760,112 @@ export function AudioPage({
() => handleClearGallery(ttsWorkflow),
[handleClearGallery, ttsWorkflow],
);
+ // Switching pages mid-run would stop the run in progress; wait for it instead.
+ const runBusy = useCallback(() => {
+ if (busyRef.current === null) return false;
+ toast.info("Wait for the current audio task to finish, then send it.");
+ return true;
+ }, [busyRef]);
const handleSendToConvert = useCallback(
- (clip: AudioGalleryClip) => {
+ (clip: AudioGalleryClip, name = clip.prompt) => {
if (!transitionWorkflow("convert")) return;
useAudioConvertStore.getState().setSource({
kind: "clip",
id: clip.id,
- name: clip.prompt,
+ name,
durationS: clip.duration_s ?? null,
});
},
[transitionWorkflow],
);
const sendHandlersFor = useCallback(
- (clip: AudioGalleryClip): ClipSendHandlers => ({
- clone: () => {
- if (transitionWorkflow("clone"))
- adoptReference(clipReference({ ...clip, workflow: clipWorkflow(clip) }));
- },
- convert: () => handleSendToConvert(clip),
- transcribe: () => {
- // Switching mid-run would stop the run in progress; wait for it instead.
- if (busyRef.current !== null) {
- toast.info("Wait for the current audio task to finish, then send the clip.");
- return;
- }
- if (!transitionWorkflow("transcribe")) return;
- // The clip goes in by id, as "From history" does; the run keeps the page's own settings.
- useAudioTranscribeStore.setState({
- source: {
+ (clip: AudioGalleryClip): ClipSendHandlers => {
+ const reference = () =>
+ clipReference({ ...clip, workflow: clipWorkflow(clip) });
+ const handlers: ClipSendHandlers = {
+ clone: () => {
+ if (transitionWorkflow("clone")) adoptReference(reference());
+ },
+ edit: () => {
+ if (transitionWorkflow("edit")) adoptEditSource(reference());
+ },
+ convert: () => handleSendToConvert(clip),
+ separate: () => {
+ if (!transitionWorkflow("separate")) return;
+ useAudioSeparateStore.getState().setSource({
kind: "clip",
id: clip.id,
name: clip.prompt || "Generated clip",
durationS: clip.duration_s,
- transcript: clip.prompt || null,
- language: null,
+ });
+ },
+ transcribe: () => {
+ if (!transitionWorkflow("transcribe")) return;
+ // The clip goes in by id, as "From history" does; the run keeps the page's own settings.
+ useAudioTranscribeStore.setState({
+ source: {
+ kind: "clip",
+ id: clip.id,
+ name: clip.prompt || "Generated clip",
+ durationS: clip.duration_s,
+ transcript: clip.prompt || null,
+ language: null,
+ },
+ });
+ },
+ };
+ return Object.fromEntries(
+ clipSendTargets(clip, ttsWorkflow).map((id) => [
+ id,
+ () => {
+ if (!runBusy()) handlers[id]?.();
},
- });
- },
- }),
- [transitionWorkflow, busyRef, handleSendToConvert],
+ ]),
+ );
+ },
+ [transitionWorkflow, runBusy, handleSendToConvert, ttsWorkflow],
+ );
+ const sendTranscriptHandlersFor = useCallback(
+ (transcript: {
+ text: string;
+ source?: TranscriptSource | null;
+ duration: number | null;
+ }): ClipSendHandlers => {
+ const { text, source, duration } = transcript;
+ // Its transcript comes along, so Edit and Clone skip transcribing it again.
+ const selection = source
+ ? { ...source, durationS: duration, transcript: text, language: null }
+ : null;
+ const handlers: ClipSendHandlers = {
+ speak: () => {
+ if (transitionWorkflow("speak")) setPrompt(text);
+ },
+ edit: () => {
+ if (!(selection && transitionWorkflow("edit"))) return;
+ adoptEditSource(selection);
+ useAudioEditStore
+ .getState()
+ .setTranscript(text.trim(), selection.id);
+ },
+ clone: () => {
+ if (!(selection && transitionWorkflow("clone"))) return;
+ adoptReference(selection);
+ // Replaces typed text for an earlier reference; blank past 30 s (Clone sends only the start).
+ useAudioCloneStore
+ .getState()
+ .applyTranscript(selection, referenceTranscript(selection).trim());
+ },
+ };
+ return Object.fromEntries(
+ transcriptSendTargets({ text, source, duration }).map((id) => [
+ id,
+ () => {
+ if (!runBusy()) handlers[id]?.();
+ },
+ ]),
+ );
+ },
+ [transitionWorkflow, runBusy, setPrompt],
);
const handleUseTextAgain = useCallback(
(clip: AudioGalleryClip) => {
@@ -803,8 +878,35 @@ export function AudioPage({
},
[transitionWorkflow, setPrompt],
);
+ const [voiceClip, setVoiceClip] = useState<{
+ id: string;
+ name: string;
+ transcript: string;
+ } | null>(null);
+ const [savingVoice, setSavingVoice] = useState(false);
+ useEffect(() => {
+ if (!active) return;
+ return () => setSavingVoice(false);
+ }, [active]);
+ const handleSaveVoice = useCallback((clip: AudioGalleryClip) => {
+ setVoiceClip({
+ id: clip.id,
+ // A clip's text is no name; the voice it was made from is, when it has one.
+ name: (clip.reference_name ?? "").replace(/\.[a-z0-9]{2,4}$/i, ""),
+ transcript: referenceTranscript(
+ clipReference({ ...clip, workflow: clipWorkflow(clip) }),
+ ),
+ });
+ setSavingVoice(true);
+ }, []);
const handleSendStem = useCallback(
async (target: SendTarget, clip: AudioGalleryClip, name: string) => {
+ if (target.workflow === "voice") {
+ setVoiceClip({ id: clip.id, name, transcript: "" });
+ setSavingVoice(true);
+ return;
+ }
+ if (runBusy()) return;
if (target.workflow === "clone") {
if (!transitionWorkflow("clone")) return;
// Adopted like any other clip, so the old reference's transcript does not stay attached.
@@ -816,12 +918,15 @@ export function AudioPage({
});
return;
}
+ if (target.workflow === "convert") {
+ handleSendToConvert(clip, name);
+ return;
+ }
+ if (target.workflow === "music") {
+ sendClipToMusic(clip, "edit", name);
+ return;
+ }
if (target.workflow === "transcribe") {
- // As in sendHandlersFor: switching mid-run would stop the run in progress.
- if (busyRef.current !== null) {
- toast.info("Wait for the current audio task to finish, then send the stem.");
- return;
- }
if (!transitionWorkflow("transcribe")) return;
// A stem is a history clip, so it goes in by id like any other.
useAudioTranscribeStore.setState({
@@ -836,7 +941,7 @@ export function AudioPage({
});
}
},
- [transitionWorkflow, busyRef],
+ [transitionWorkflow, runBusy, handleSendToConvert],
);
const pageModelLoaded =
@@ -1429,6 +1534,26 @@ export function AudioPage({
<MediaRailResizeHandle kind="audio" placement="page" className="hidden @[50rem]:block" />
{/* Portals to body, and this page stays mounted off-route, so gate it like the composer. */}
{active && <GuidedTour {...tour.tourProps} />}
+ <SaveVoiceDialog
+ open={active && savingVoice}
+ onOpenChange={setSavingVoice}
+ mode="create"
+ initial={{
+ name: voiceClip?.name ?? "",
+ transcript: voiceClip?.transcript ?? "",
+ language: "",
+ }}
+ onSubmit={async (details) => {
+ if (!voiceClip) return;
+ const voice = await useAudioVoicesStore.getState().save({
+ source: { clip_id: voiceClip.id },
+ name: details.name,
+ transcript: details.transcript || null,
+ language: details.language || null,
+ });
+ toast.success(`Saved ${voice.name}. Pick it under Saved voice next time.`);
+ }}
+ />
{/* Keep the tabs centered over the preview at every width. The model rail holds at its
(draggable) width when space permits and shrinks only to preserve the controls. */}
<div className="pointer-events-none relative z-40 grid h-[calc(48px*var(--ui-space-scale,1))] shrink-0 grid-cols-[minmax(0,var(--media-rail-width,calc(408px*var(--ui-space-scale,1))))_minmax(13rem,1fr)] @max-[30rem]:grid-cols-[minmax(0,1fr)_auto]">
@@ -1817,6 +1942,7 @@ export function AudioPage({
speakerNames={speakerNames}
renameSpeaker={renameSpeaker}
selectRecord={selectRecord}
+ sendHandlersFor={sendTranscriptHandlersFor}
/>
<output aria-live="polite" aria-atomic="true" className="sr-only">
{transcribeAnnouncement}
@@ -1829,6 +1955,7 @@ export function AudioPage({
workflow: ttsWorkflow,
peaksById,
sendHandlersFor,
+ onSaveVoice: handleSaveVoice,
pending:
busy === "generating" && generationPresentation
? {
diff --git a/studio/frontend/src/features/audio/components/clip-card.tsx b/studio/frontend/src/features/audio/components/clip-card.tsx
index c8597c27643..88a119d06dd 100644
--- a/studio/frontend/src/features/audio/components/clip-card.tsx
+++ b/studio/frontend/src/features/audio/components/clip-card.tsx
@@ -19,17 +19,17 @@ import { Waveform } from "./waveform";
export type ClipSendHandlers = Partial<Record<AudioWorkflowId, () => void>>;
-export function ClipSendToMenu({
- current,
- handlers,
-}: {
- current: AudioWorkflowId;
- handlers: ClipSendHandlers;
-}) {
- const targets = AUDIO_WORKFLOWS.filter(
- (tab) => tab.id !== current && handlers[tab.id],
- );
- if (targets.length === 0) return null;
+export function SendToItems({ handlers }: { handlers: ClipSendHandlers }) {
+ return AUDIO_WORKFLOWS.filter((tab) => handlers[tab.id]).map((tab) => (
+ <DropdownMenuItem key={tab.id} onClick={handlers[tab.id]}>
+ <HugeiconsIcon icon={tab.icon} strokeWidth={1.75} className="size-icon" />
+ {tab.label}
+ </DropdownMenuItem>
+ ));
+}
+
+export function ClipSendToMenu({ handlers }: { handlers: ClipSendHandlers }) {
+ if (!Object.values(handlers).some(Boolean)) return null;
return (
<DropdownMenuSub>
<DropdownMenuSubTrigger>
@@ -41,16 +41,7 @@ export function ClipSendToMenu({
Send to
</DropdownMenuSubTrigger>
<DropdownMenuSubContent>
- {targets.map((tab) => (
- <DropdownMenuItem key={tab.id} onClick={handlers[tab.id]}>
- <HugeiconsIcon
- icon={tab.icon}
- strokeWidth={1.75}
- className="size-icon"
- />
- {tab.label}
- </DropdownMenuItem>
- ))}
+ <SendToItems handlers={handlers} />
</DropdownMenuSubContent>
</DropdownMenuSub>
);
@@ -58,7 +49,7 @@ export function ClipSendToMenu({
const CARD_CLASS = "w-full max-w-xl gap-3 px-5 py-4";
-/** Never autoplays; a clip a run just made takes focus on its play button once. */
+/** never autoplays; a new clip focuses its play button once. */
export function ClipCard({
title,
model,
diff --git a/studio/frontend/src/features/audio/components/music-send-to.tsx b/studio/frontend/src/features/audio/components/music-send-to.tsx
index 4066d5045d0..9e9d0329a75 100644
--- a/studio/frontend/src/features/audio/components/music-send-to.tsx
+++ b/studio/frontend/src/features/audio/components/music-send-to.tsx
@@ -13,7 +13,7 @@ import { useAudioMusicStore } from "../stores/audio-music-store";
import { useAudioWorkspaceStore } from "../stores/audio-workspace-store";
import { clipWorkflow } from "../workflows";
-// History is per page, so these only ever show on Music itself.
+// history is scoped per page, so these actions appear only on Music.
const SEND_LABEL: Record<SendAction, string> = {
edit: "Edit",
extend: "Extend",
@@ -27,12 +27,13 @@ const SEND_ICON = {
export function sendClipToMusic(
clip: AudioGalleryClip,
action: SendAction,
+ name = clip.prompt || "Generated clip",
): void {
useAudioMusicStore.getState().pushClipToEdit(
{
kind: "clip",
id: clip.id,
- name: clip.prompt || "Generated clip",
+ name,
durationS: clip.duration_s,
transcript: null,
language: null,
diff --git a/studio/frontend/src/features/audio/hooks/use-music-generation.ts b/studio/frontend/src/features/audio/hooks/use-music-generation.ts
index 3d08860c553..a66db0b48bc 100644
--- a/studio/frontend/src/features/audio/hooks/use-music-generation.ts
+++ b/studio/frontend/src/features/audio/hooks/use-music-generation.ts
@@ -20,6 +20,7 @@ import {
musicBlocker,
reloadNotice,
variationsFor,
+ withWaitingEdit,
} from "../music/music-policy";
import { parseMusicCapabilities } from "../music/music-types";
import type {
@@ -112,16 +113,17 @@ export function useMusicGeneration({
modelName: string | null;
musicLoaded: boolean;
}) {
- const capabilities = useMemo(
- () =>
- parseMusicCapabilities(status?.audio_music) ??
- (musicLoaded ? null : MUSIC_PREVIEW_CAPABILITIES),
- [status?.audio_music, musicLoaded],
- );
const pickedMode = useAudioMusicStore((state) => state.mode);
const song = useAudioMusicStore((state) => state.song);
const sfx = useAudioMusicStore((state) => state.sfx);
const edit = useAudioMusicStore((state) => state.edit);
+ const editWaiting = edit.source !== null;
+ const capabilities = useMemo(() => {
+ const loaded =
+ parseMusicCapabilities(status?.audio_music) ??
+ (musicLoaded ? null : MUSIC_PREVIEW_CAPABILITIES);
+ return loaded && withWaitingEdit(loaded, editWaiting);
+ }, [status?.audio_music, musicLoaded, editWaiting]);
const rule = capabilities
? effectiveMusicMode(capabilities, pickedMode)
: null;
diff --git a/studio/frontend/src/features/audio/music/music-policy.ts b/studio/frontend/src/features/audio/music/music-policy.ts
index eec8dd68611..1e4e25a94a9 100644
--- a/studio/frontend/src/features/audio/music/music-policy.ts
+++ b/studio/frontend/src/features/audio/music/music-policy.ts
@@ -55,6 +55,17 @@ export function effectiveMusicMode(
);
}
+/** A clip sent to edit stays in view even when the model cannot edit; Edit then says to load one that can. */
+export function withWaitingEdit(
+ capabilities: MusicCapabilities,
+ editWaiting: boolean,
+): MusicCapabilities {
+ if (!editWaiting || capabilities.modes.some((rule) => rule.id === "edit")) {
+ return capabilities;
+ }
+ return { modes: [...capabilities.modes, { id: "edit" }] };
+}
+
const SECTION_NAMES = [
"Intro",
"Verse",
diff --git a/studio/frontend/src/features/audio/pages/separate-page.tsx b/studio/frontend/src/features/audio/pages/separate-page.tsx
index bbd138873ff..5f9a3458796 100644
--- a/studio/frontend/src/features/audio/pages/separate-page.tsx
+++ b/studio/frontend/src/features/audio/pages/separate-page.tsx
@@ -35,6 +35,7 @@ import {
type SeparateGeneration,
} from "../hooks/use-separate-generation";
import { useStemSources } from "../hooks/use-stem-sources";
+import { STEM_SEND_TARGETS } from "../send-targets";
import { saveAudio } from "../save-audio";
import {
SEPARATE_MAX_SECONDS,
@@ -281,12 +282,7 @@ function downloadGroup(group: SeparationGroup) {
return saveAudio(stemZipName(group.title), null, () => zipStems(files));
}
-const SEND_TARGETS: readonly SendTarget[] = [
- { id: "transcribe", workflow: "transcribe", label: "Transcribe" },
- { id: "clone", workflow: "clone", label: "Clone (as reference)" },
-];
-
-/** Keyed by group so each one starts from its own clock. */
+/** keyed by group so each mixer starts from its own clock. */
function SelectedSeparation({
group,
autoFocus,
@@ -331,7 +327,7 @@ function SelectedSeparation({
stems={stems}
autoFocus={autoFocus}
active={active}
- sendTargets={SEND_TARGETS}
+ sendTargets={STEM_SEND_TARGETS}
onDownloadStem={(clipId) => {
const clip = group.stems.find((item) => item.id === clipId);
const src = sources.srcById[clipId];
diff --git a/studio/frontend/src/features/audio/pages/transcribe-page.tsx b/studio/frontend/src/features/audio/pages/transcribe-page.tsx
index ef826a71b8e..a86b7341266 100644
--- a/studio/frontend/src/features/audio/pages/transcribe-page.tsx
+++ b/studio/frontend/src/features/audio/pages/transcribe-page.tsx
@@ -20,6 +20,7 @@ import { cn } from "@/lib/utils";
import {
Copy01Icon,
Download01Icon,
+ SentIcon,
StopIcon,
} from "@hugeicons/core-free-icons";
import { HugeiconsIcon } from "@hugeicons/react";
@@ -32,9 +33,14 @@ import {
AudioSourceInput,
type AudioSourceInputHandle,
} from "../components/audio-source-input";
+import { SendToItems } from "../components/clip-card";
import { Field } from "../components/field";
import { TranscriptView } from "../components/transcript-view";
-import { TranscriptExportItems, TranscriptGallery } from "../transcript-gallery";
+import {
+ TranscriptExportItems,
+ TranscriptGallery,
+ type TranscriptSendHandlers,
+} from "../transcript-gallery";
import { TranscriptionProgress } from "../transcription-progress";
import type { AudioHostState } from "../hooks/audio-host-state";
import type { AudioSourceStatus } from "../hooks/audio-source-state";
@@ -237,6 +243,7 @@ export function TranscribeOutput({
speakerNames,
renameSpeaker,
selectRecord,
+ sendHandlersFor,
}: Pick<
Transcription,
| "transcriptionStartedAt"
@@ -260,9 +267,16 @@ export function TranscribeOutput({
| "renameSpeaker"
| "selectRecord"
> &
- Pick<AudioHostState, "busy" | "active" | "mode">) {
+ Pick<AudioHostState, "busy" | "active" | "mode"> & {
+ sendHandlersFor: TranscriptSendHandlers;
+ }) {
const language = transcriptDetails.language ?? transcriptRecord?.language;
const duration = transcriptDetails.duration ?? transcriptRecord?.duration;
+ const sendHandlers = sendHandlersFor({
+ text: transcript,
+ source: transcriptDetails.source ?? transcriptRecord?.source,
+ duration: duration ?? null,
+ });
return (
<>
<div className="hover-scrollbar flex min-h-0 flex-1 flex-col gap-3 overflow-y-auto">
@@ -326,6 +340,19 @@ export function TranscribeOutput({
/>
</DropdownMenuContent>
</DropdownMenu>
+ {Object.keys(sendHandlers).length > 0 ? (
+ <DropdownMenu>
+ <DropdownMenuTrigger asChild={true}>
+ <Button variant="secondary" size="sm">
+ <HugeiconsIcon icon={SentIcon} className="mr-2 size-3.5" />
+ Send to
+ </Button>
+ </DropdownMenuTrigger>
+ <DropdownMenuContent align="start">
+ <SendToItems handlers={sendHandlers} />
+ </DropdownMenuContent>
+ </DropdownMenu>
+ ) : null}
</div>
<div className="flex min-w-0 flex-wrap items-center gap-x-2 text-ui-11p5 text-muted-foreground">
<span className="min-w-0 truncate">{transcribedName}</span>
@@ -370,6 +397,7 @@ export function TranscribeOutput({
latest={transcriptRecord}
canSelect={confirmTranscriptReplacement}
onSelect={selectRecord}
+ sendHandlersFor={sendHandlersFor}
onDelete={(ids) => {
if (
transcriptRecord &&
diff --git a/studio/frontend/src/features/audio/pages/tts-workspace.tsx b/studio/frontend/src/features/audio/pages/tts-workspace.tsx
index 3cc240bdbbf..e322ee2348c 100644
--- a/studio/frontend/src/features/audio/pages/tts-workspace.tsx
+++ b/studio/frontend/src/features/audio/pages/tts-workspace.tsx
@@ -21,6 +21,7 @@ import { cn } from "@/lib/utils";
import {
AudioWave01Icon,
Copy01Icon,
+ FloppyDiskIcon,
SparklesIcon,
StopIcon,
} from "@hugeicons/core-free-icons";
@@ -445,7 +446,6 @@ function ClipBadge({ text }: { text: string | null | undefined }) {
}
export function TtsOutput({
- workflow,
clips,
selectedClip,
selectedClipSrc,
@@ -470,6 +470,7 @@ export function TtsOutput({
onUseTextAgain,
handleCopyPrompt,
sendHandlersFor,
+ onSaveVoice,
pending,
freshClipId,
onFreshClipFocused,
@@ -505,6 +506,7 @@ export function TtsOutput({
handleClearGallery: () => Promise<void>;
onUseTextAgain: (clip: AudioGalleryClip) => void;
sendHandlersFor: (clip: AudioGalleryClip) => ClipSendHandlers;
+ onSaveVoice: (clip: AudioGalleryClip) => void;
pending: {
title: string;
status: string;
@@ -517,7 +519,7 @@ export function TtsOutput({
freshClipId: string | null;
onFreshClipFocused: () => void;
announcement: string;
- /** A short tag after a clip's text, such as the voice a clone used. */
+ /** short tag after a clip's text, such as the voice a clone used. */
clipBadge?: (
clip: AudioGalleryClip,
place: "selected" | "history",
@@ -574,7 +576,15 @@ export function TtsOutput({
onAddToProject={(projectId) => addAudioClipToProject(clip.id, projectId)}
leadingItems={
<>
- <ClipSendToMenu current={workflow} handlers={sendHandlersFor(clip)} />
+ <ClipSendToMenu handlers={sendHandlersFor(clip)} />
+ <DropdownMenuItem onClick={() => onSaveVoice(clip)}>
+ <HugeiconsIcon
+ icon={FloppyDiskIcon}
+ strokeWidth={1.75}
+ className="size-icon"
+ />
+ Save as voice…
+ </DropdownMenuItem>
<DropdownMenuItem onClick={() => onUseTextAgain(clip)}>
<HugeiconsIcon
icon={SparklesIcon}
diff --git a/studio/frontend/src/features/audio/send-targets.ts b/studio/frontend/src/features/audio/send-targets.ts
new file mode 100644
index 00000000000..d6efced4cbe
--- /dev/null
+++ b/studio/frontend/src/features/audio/send-targets.ts
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+// free of app imports so the node test runner can load it directly.
+
+import {
+ EDIT_SOURCE_MAX_SECONDS,
+ EDIT_SOURCE_SLACK_SECONDS,
+} from "./audio-run-request";
+import type { SendTarget } from "./components/stem-mixer-types";
+import type { TranscriptSource } from "./transcript-model";
+import { type AudioWorkflowId, clipWorkflow } from "./workflows";
+
+const SPEECH_WORKFLOWS: ReadonlySet<string> = new Set([
+ "speak",
+ "clone",
+ "edit",
+ "convert",
+]);
+
+// edit refuses a longer recording rather than cutting it.
+function fitsEdit(durationS: number | null | undefined): boolean {
+ return (
+ durationS == null ||
+ durationS <= EDIT_SOURCE_MAX_SECONDS + EDIT_SOURCE_SLACK_SECONDS
+ );
+}
+
+/** edit accepts only speech, including its own results; other pages accept any clip. */
+export function clipSendTargets(
+ clip: {
+ workflow?: string | null;
+ audio_type?: string | null;
+ duration_s: number | null;
+ },
+ current: AudioWorkflowId,
+): AudioWorkflowId[] {
+ const targets: AudioWorkflowId[] = [
+ "clone",
+ "convert",
+ "separate",
+ "transcribe",
+ ];
+ if (SPEECH_WORKFLOWS.has(clipWorkflow(clip)) && fitsEdit(clip.duration_s)) {
+ targets.push("edit");
+ }
+ return targets.filter((id) => id !== current || id === "edit");
+}
+
+export const STEM_SEND_TARGETS: readonly SendTarget[] = [
+ { id: "clone", workflow: "clone", label: "Clone (as reference)" },
+ { id: "convert", workflow: "convert", label: "Convert" },
+ { id: "music", workflow: "music", label: "Music edit" },
+ { id: "transcribe", workflow: "transcribe", label: "Transcribe" },
+ { id: "voice", workflow: "voice", label: "Save as voice…" },
+];
+
+/** text goes to speak; edit and clone also require its source audio. */
+export function transcriptSendTargets({
+ text,
+ source,
+ duration,
+}: {
+ text: string;
+ source: TranscriptSource | null | undefined;
+ duration: number | null | undefined;
+}): AudioWorkflowId[] {
+ if (!text.trim()) return [];
+ const targets: AudioWorkflowId[] = ["speak"];
+ if (!source) return targets;
+ // edit cannot take a saved voice as its recording.
+ if (source.kind !== "voice" && fitsEdit(duration)) targets.push("edit");
+ targets.push("clone");
+ return targets;
+}
diff --git a/studio/frontend/src/features/audio/stores/audio-clone-store.ts b/studio/frontend/src/features/audio/stores/audio-clone-store.ts
index e2ccdf988bc..ee64c9d08b8 100644
--- a/studio/frontend/src/features/audio/stores/audio-clone-store.ts
+++ b/studio/frontend/src/features/audio/stores/audio-clone-store.ts
@@ -10,13 +10,13 @@ import {
const AUDIO_CLONE_STORAGE_KEY = "unsloth_audio_clone_v1";
-// Clone sends the reference's first 30 s, so a longer clip's full text would not match it.
-function referenceTranscript(next: AudioSourceSelection | null): string {
+// clone sends the reference's first 30 s, so a longer clip's full text would not match it.
+export function referenceTranscript(next: AudioSourceSelection | null): string {
if (!next || (next.durationS ?? 0) > REFERENCE_MAX_SECONDS) return "";
return next.transcript || "";
}
-/** Also holds every page's tool values, not just Clone's. */
+/** also holds every page's tool values, not just clone's. */
interface AudioCloneState {
reference: AudioSourceSelection | null;
referenceText: string;
diff --git a/studio/frontend/src/features/audio/transcript-gallery.tsx b/studio/frontend/src/features/audio/transcript-gallery.tsx
index c69c5e30fcf..89e4497c528 100644
--- a/studio/frontend/src/features/audio/transcript-gallery.tsx
+++ b/studio/frontend/src/features/audio/transcript-gallery.tsx
@@ -30,12 +30,20 @@ import { copyToClipboard } from "@/lib/copy-to-clipboard";
import { toast } from "@/lib/toast";
import { cn } from "@/lib/utils";
import { archiveTranscript, deleteTranscript, listTranscripts } from "./api";
+import {
+ type ClipSendHandlers,
+ ClipSendToMenu,
+} from "./components/clip-card";
import { getTranscript } from "./transcribe-api";
import {
TRANSCRIPT_EXPORT_FORMATS,
type TranscriptExportFormat,
} from "./transcript-export";
-import { detailsFrom, formatTimestamp } from "./transcript-model";
+import {
+ type TranscriptSource,
+ detailsFrom,
+ formatTimestamp,
+} from "./transcript-model";
import type { TranscriptRecord } from "./transcript-stream";
export function TranscriptExportItems({
@@ -98,6 +106,12 @@ function recordBadge(record: TranscriptRecord): string | null {
return (record.segment_count ?? 0) > 0 ? "Timestamps" : null;
}
+export type TranscriptSendHandlers = (transcript: {
+ text: string;
+ source?: TranscriptSource | null;
+ duration: number | null;
+}) => ClipSendHandlers;
+
export function TranscriptGallery({
active,
currentId,
@@ -106,6 +120,7 @@ export function TranscriptGallery({
onDelete,
canSelect,
autoSelect,
+ sendHandlersFor,
}: {
active: boolean;
autoSelect: boolean;
@@ -114,6 +129,7 @@ export function TranscriptGallery({
onSelect: (record: TranscriptRecord) => void;
onDelete: (ids: string[] | null) => void;
canSelect: () => boolean;
+ sendHandlersFor: TranscriptSendHandlers;
}) {
const [records, setRecords] = useState<TranscriptRecord[]>([]);
const [archived, setArchived] = useState(false);
@@ -306,6 +322,7 @@ export function TranscriptGallery({
</Button>
</DropdownMenuTrigger>
<DropdownMenuContent align="end">
+ <ClipSendToMenu handlers={sendHandlersFor(record)} />
<TranscriptExportItems
timed={(record.segment_count ?? 0) > 0}
label={(format) => (
diff --git a/studio/frontend/tests/audio-clone-store.test.ts b/studio/frontend/tests/audio-clone-store.test.ts
index 5a4e1f00e2a..c3e0549c65b 100644
--- a/studio/frontend/tests/audio-clone-store.test.ts
+++ b/studio/frontend/tests/audio-clone-store.test.ts
@@ -228,3 +228,37 @@ test("a reference longer than the 30 s cut brings no transcript, since only 30 s
});
assert.equal(store().referenceText, "Hi.");
});
+
+test("a transcript sent with its recording replaces words typed for another", () => {
+ const store = () => useAudioCloneStore.getState();
+ const sent = {
+ kind: "clip" as const,
+ id: "sent1",
+ name: "take.wav",
+ durationS: 3,
+ transcript: "the sent words",
+ };
+ store().setReferenceText("typed for the old clip");
+ store().adoptReference(sent);
+ assert.equal(store().referenceText, "typed for the old clip");
+ store().applyTranscript(sent, "the sent words");
+ assert.equal(store().referenceText, "the sent words");
+});
+
+test("a sent transcript of a recording over 30 s leaves Clone's text blank", async () => {
+ const { referenceTranscript } = await import(
+ "../src/features/audio/stores/audio-clone-store.ts"
+ );
+ const store = () => useAudioCloneStore.getState();
+ const long = {
+ kind: "clip" as const,
+ id: "long1",
+ name: "talk.wav",
+ durationS: 45,
+ transcript: "forty five seconds of words",
+ };
+ store().setReferenceText("typed for the old clip");
+ store().adoptReference(long);
+ store().applyTranscript(long, referenceTranscript(long));
+ assert.equal(store().referenceText, "");
+});
diff --git a/studio/frontend/tests/audio-convert-policy.test.ts b/studio/frontend/tests/audio-convert-policy.test.ts
index 6cd016ef022..74f0e139a81 100644
--- a/studio/frontend/tests/audio-convert-policy.test.ts
+++ b/studio/frontend/tests/audio-convert-policy.test.ts
@@ -372,7 +372,7 @@ test("Use again re-uploads a conversion's kept source instead of its expiring up
const SEND_TO_HAS_CONVERT = /convert: \(\) => handleSendToConvert\(clip\),/;
const SEND_TO_CLONE_KEEPS_LABELS_OUT =
- /adoptReference\(clipReference\(\{ \.\.\.clip, workflow: clipWorkflow\(clip\) \}\)\)/;
+ /const reference = \(\) =>\s*clipReference\(\{ \.\.\.clip, workflow: clipWorkflow\(clip\) \}\);\s*const handlers: ClipSendHandlers = \{\s*clone: \(\) => \{\s*if \(transitionWorkflow\("clone"\)\) adoptReference\(reference\(\)\);/;
const WORKSPACE_FOCUSES_A_FRESH_CUSTOM_PLAYER =
/renderPlayer\(\s*selectedClip,\s*selectedClipSrc,\s*selectedClip\.id === freshClipId \? focusFreshClip : undefined,?\s*\)/;
diff --git a/studio/frontend/tests/audio-edit-store.test.ts b/studio/frontend/tests/audio-edit-store.test.ts
index c9868524f80..6d564d9ac85 100644
--- a/studio/frontend/tests/audio-edit-store.test.ts
+++ b/studio/frontend/tests/audio-edit-store.test.ts
@@ -54,6 +54,11 @@ test("② follows ① until the user types in it, and Reset puts it back", () =>
state().setEdited("It was a rainy day.");
state().setTranscript("It was a fine day!");
assert.deepEqual(draft(), ["It was a rainy day.", "in2"]);
+ state().setTranscript("A newer saved transcript.", "in2");
+ assert.deepEqual(
+ [state().transcript, ...draft()],
+ ["A newer saved transcript.", "It was a rainy day.", "in2"],
+ );
state().setDelivery({ pitchSteps: 4 });
assert.deepEqual(state().delivery, { speed: 1.5, pitchSteps: 4 });
const { state: saved, version } = JSON.parse(store.get(KEY) ?? "{}");
diff --git a/studio/frontend/tests/audio-music-policy.test.ts b/studio/frontend/tests/audio-music-policy.test.ts
index 9c9b6fd568b..a8bdbd3ea59 100644
--- a/studio/frontend/tests/audio-music-policy.test.ts
+++ b/studio/frontend/tests/audio-music-policy.test.ts
@@ -15,6 +15,7 @@ const {
buildMusicRunRequest,
durationLabel,
effectiveMusicMode,
+ withWaitingEdit,
musicModeHint,
musicModeLabel,
insertSectionTag,
@@ -405,3 +406,23 @@ test("instrumental-only models say so in the mode label and length wording", ()
"Length (seconds, approximate)",
);
});
+
+test("a clip waiting in Edit keeps the Edit tab, even on a model that cannot edit", () => {
+ const songOnly = { modes: [{ id: "song" as const }] };
+ const waiting = withWaitingEdit(songOnly, true);
+ assert.deepEqual(
+ waiting.modes.map((rule) => rule.id),
+ ["song", "edit"],
+ );
+ // The added mode has no actions, so the page says the model cannot edit and Generate stays blocked.
+ assert.equal(effectiveMusicMode(waiting, "edit").actions, undefined);
+ assert.equal(effectiveMusicMode(songOnly, "edit").id, "song");
+ assert.equal(withWaitingEdit(songOnly, false), songOnly);
+ const editing = {
+ modes: [
+ { id: "song" as const },
+ { id: "edit" as const, actions: ["repaint" as const] },
+ ],
+ };
+ assert.equal(withWaitingEdit(editing, true), editing);
+});
diff --git a/studio/frontend/tests/audio-send-targets.test.ts b/studio/frontend/tests/audio-send-targets.test.ts
new file mode 100644
index 00000000000..1bf972fe396
--- /dev/null
+++ b/studio/frontend/tests/audio-send-targets.test.ts
@@ -0,0 +1,148 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { readSrc, registerBundlerResolver } from "./helpers/kit.ts";
+
+registerBundlerResolver();
+const { STEM_SEND_TARGETS, clipSendTargets, transcriptSendTargets } =
+ await import("../src/features/audio/send-targets.ts");
+
+const clip = (workflow: string, duration_s: number | null = 4) => ({
+ workflow,
+ duration_s,
+});
+
+test("speech clips can be edited, and Edit takes its own results again", () => {
+ assert.deepEqual(clipSendTargets(clip("speak"), "speak"), [
+ "clone",
+ "convert",
+ "separate",
+ "transcribe",
+ "edit",
+ ]);
+ assert.ok(clipSendTargets(clip("clone"), "clone").includes("edit"));
+ assert.ok(clipSendTargets(clip("convert"), "convert").includes("edit"));
+ assert.deepEqual(clipSendTargets(clip("edit"), "edit"), [
+ "clone",
+ "convert",
+ "separate",
+ "transcribe",
+ "edit",
+ ]);
+ assert.ok(!clipSendTargets(clip("clone"), "clone").includes("clone"));
+ assert.ok(!clipSendTargets(clip("convert"), "convert").includes("convert"));
+});
+
+test("music is split into stems but not sent to speech Edit", () => {
+ assert.deepEqual(clipSendTargets(clip("music", 30), "music"), [
+ "clone",
+ "convert",
+ "separate",
+ "transcribe",
+ ]);
+ // older clips carry only an audio type
+ assert.ok(
+ !clipSendTargets(
+ { audio_type: "minimax_music3", duration_s: 10 },
+ "music",
+ ).includes("edit"),
+ );
+});
+
+test("a clip longer than Edit takes is not offered to Edit", () => {
+ assert.ok(clipSendTargets(clip("speak", 30.05), "speak").includes("edit"));
+ assert.ok(!clipSendTargets(clip("speak", 31), "speak").includes("edit"));
+ // an unknown length is left to Edit's own check
+ assert.ok(clipSendTargets(clip("speak", null), "speak").includes("edit"));
+});
+
+test("stems go to Clone, Convert, Music edit and Transcribe, or become a voice", () => {
+ assert.deepEqual(
+ STEM_SEND_TARGETS.map((target) => target.workflow),
+ ["clone", "convert", "music", "transcribe", "voice"],
+ );
+ assert.equal(
+ STEM_SEND_TARGETS.find((target) => target.workflow === "music")?.label,
+ "Music edit",
+ );
+});
+
+test("a transcript's text goes to Speak; Edit and Clone also need its audio", () => {
+ const input = { kind: "input" as const, id: "a1", name: "take.wav" };
+ assert.deepEqual(
+ transcriptSendTargets({ text: "hello", source: input, duration: 12 }),
+ ["speak", "edit", "clone"],
+ );
+ assert.deepEqual(
+ transcriptSendTargets({ text: "hello", source: null, duration: 12 }),
+ ["speak"],
+ );
+ assert.deepEqual(
+ transcriptSendTargets({
+ text: "hello",
+ source: { kind: "voice", id: "v1", name: "Narrator" },
+ duration: 8,
+ }),
+ ["speak", "clone"],
+ );
+ assert.deepEqual(
+ transcriptSendTargets({ text: "hello", source: input, duration: 95 }),
+ ["speak", "clone"],
+ );
+ assert.deepEqual(
+ transcriptSendTargets({ text: " ", source: input, duration: 3 }),
+ [],
+ );
+});
+
+test("the Audio page wires every send, and waits out a running task first", () => {
+ const host = readSrc("features/audio/audio-page.tsx");
+ assert.match(host, /clipSendTargets\(clip, ttsWorkflow\)/);
+ assert.match(host, /transcriptSendTargets\(\{ text, source, duration \}\)/);
+ assert.match(host, /if \(!runBusy\(\)\) handlers\[id\]\?\.\(\)/);
+ assert.match(
+ host,
+ /if \(transitionWorkflow\("edit"\)\) adoptEditSource\(reference\(\)\)/,
+ );
+ assert.match(host, /useAudioSeparateStore\.getState\(\)\.setSource\(\{/);
+ assert.match(host, /sendClipToMusic\(clip, "edit", name\)/);
+ const music = readSrc("features/audio/hooks/use-music-generation.ts");
+ assert.match(
+ music,
+ /return loaded && withWaitingEdit\(loaded, editWaiting\);/,
+ );
+ assert.match(host, /source: \{ clip_id: voiceClip\.id \}/);
+ assert.match(host, /<SaveVoiceDialog\s+open=\{active && savingVoice\}/);
+ assert.match(
+ host,
+ /useEffect\(\(\) => \{\s*if \(!active\) return;\s*return \(\) => setSavingVoice\(false\);\s*\}, \[active\]\);/,
+ );
+ const transcriptSend = host.slice(
+ host.indexOf("const sendTranscriptHandlersFor"),
+ host.indexOf("const handleUseTextAgain"),
+ );
+ assert.doesNotMatch(transcriptSend, /runTranscription|handleTranscribe/);
+ assert.match(
+ transcriptSend,
+ /adoptEditSource\(selection\);\s*useAudioEditStore\s*\.getState\(\)\s*\.setTranscript\(text\.trim\(\), selection\.id\);/,
+ );
+ assert.match(
+ transcriptSend,
+ /adoptReference\(selection\);[^]*?useAudioCloneStore\s*\.getState\(\)\s*\.applyTranscript\(selection, referenceTranscript\(selection\)\.trim\(\)\);/,
+ );
+
+ const menu = readSrc("features/audio/pages/tts-workspace.tsx");
+ assert.match(menu, /onClick=\{\(\) => onSaveVoice\(clip\)\}/);
+ const transcribe = readSrc("features/audio/pages/transcribe-page.tsx");
+ assert.match(transcribe, /<SendToItems handlers=\{sendHandlers\} \/>/);
+ const gallery = readSrc("features/audio/transcript-gallery.tsx");
+ assert.match(
+ gallery,
+ /<ClipSendToMenu handlers=\{sendHandlersFor\(record\)\} \/>/,
+ );
+ const separate = readSrc("features/audio/pages/separate-page.tsx");
+ assert.match(separate, /sendTargets=\{STEM_SEND_TARGETS\}/);
+});
diff --git a/studio/frontend/tests/audio-separate.test.ts b/studio/frontend/tests/audio-separate.test.ts
index 4f735a20b34..285036a23d0 100644
--- a/studio/frontend/tests/audio-separate.test.ts
+++ b/studio/frontend/tests/audio-separate.test.ts
@@ -281,32 +281,28 @@ test("a loaded separation model is a main-slot audio model", () => {
test("host and page wiring for Separate", () => {
const host = readSrc("features/audio/audio-page.tsx");
- // Leaving Audio stops a Separate recording and releases the mic, as on Clone.
assert.match(
host,
/ttsWorkflow === "separate" \? \(\s*<AudioActiveProvider value=\{active\}>\s*<SeparateRail/,
);
assert.match(host, /ttsWorkflow === "separate"\s*\? separate\.blocker/);
assert.match(host, /reason: "The loaded model separates audio\."/);
- // Trained speech checkpoints are not separation models: only Speak and Music offer them.
assert.match(
host,
/ttsWorkflow !== "speak" && ttsWorkflow !== "music"\s*\? \[\]/,
);
- // A stem sent to Transcribe mid-run would be dropped when the run stops.
+ // switching to Transcribe mid-run would stop the run and drop the stem
assert.match(
host,
- /target\.workflow === "transcribe"\) \{[\s\S]*?busyRef\.current !== null[\s\S]*?return;[\s\S]*?if \(!transitionWorkflow\("transcribe"\)\) return;[\s\S]*?useAudioTranscribeStore\.setState\(\{\s*source: \{\s*kind: "clip",\s*id: clip\.id/,
+ /const handleSendStem = useCallback\([\s\S]*?if \(runBusy\(\)\) return;[\s\S]*?target\.workflow === "transcribe"\) \{\s*if \(!transitionWorkflow\("transcribe"\)\) return;[\s\S]*?useAudioTranscribeStore\.setState\(\{\s*source: \{\s*kind: "clip",\s*id: clip\.id/,
);
- // A separation that a refresh missed falls back on the Separate page.
const generation = readSrc("features/audio/hooks/use-separate-generation.ts");
assert.match(generation, /showRunResult\(\{[^}]*workflow: "separate"/);
const page = readSrc("features/audio/pages/separate-page.tsx");
assert.match(page, /allowSavedVoice=\{false\}/);
assert.match(page, /expiredMessage=\{SEPARATE_TRACK_EXPIRED_MESSAGE\}/);
assert.match(page, /maxRecordSeconds=\{SEPARATE_MAX_SECONDS - 1\}/);
- // The mixer's own sources pin the group; a failed fetch retries instead of loading forever,
- // and a stem that cannot load is marked so the mixer plays the others.
+ // mixer sources pin the group; failed fetches retry, while unavailable stems are skipped
assert.match(page, /useStemSources\(inputs, attempt\)/);
assert.match(page, /sources\.failedIds\.length > 0/);
assert.match(page, /setAttempt\(\(n\) => n \+ 1\)/);
diff --git a/studio/frontend/tests/audio-workspace-split.test.ts b/studio/frontend/tests/audio-workspace-split.test.ts
index d056f4fbd68..2f3b03dcc59 100644
--- a/studio/frontend/tests/audio-workspace-split.test.ts
+++ b/studio/frontend/tests/audio-workspace-split.test.ts
@@ -133,7 +133,6 @@ test("results show as a clip card with a waveform, never autoplaying", () => {
assert.doesNotMatch(output + card, /autoPlay/);
assert.match(output, /<ClipCard\s+\/\/[^\n]*\n\s*key=\{selectedClip\.id\}/);
assert.match(card, /<Waveform\s+peaks=\{peaks\}/);
- // A run in progress stands where its clip will appear, with Stop, and no second live region.
assert.match(
output,
/\{pending \? \(\s*<PendingClipCard \{\.\.\.pending\} \/>/,
@@ -146,12 +145,10 @@ test("results show as a clip card with a waveform, never autoplaying", () => {
assert.doesNotMatch(card, /aria-live/);
});
-test("Send to lists the other Audio pages from the shared workflow list", () => {
+test("Send to lists Audio pages from the shared workflow list", () => {
const card = readSrc("features/audio/components/clip-card.tsx");
- assert.match(
- card,
- /AUDIO_WORKFLOWS\.filter\(\s*\(tab\) => tab\.id !== current && handlers\[tab\.id\],?\s*\)/,
- );
+ // send-targets decides eligibility while the menu preserves page order
+ assert.match(card, /AUDIO_WORKFLOWS\.filter\(\(tab\) => handlers\[tab\.id\]\)/);
assert.match(
host,
/if \(!transitionWorkflow\("transcribe"\)\) return;[\s\S]{0,200}?useAudioTranscribeStore\.setState\(\{\s*source: \{\s*kind: "clip",\s*id: clip\.id,/,
@@ -185,11 +182,15 @@ test("auto-selecting a page's history keeps another page's unsaved clip", () =>
});
test("Send to waits for a running task instead of stopping it and dropping the clip", () => {
- // A switch mid-run stops the run, but the page stays busy until the stopped run settles, so the
- // transcription was refused and the clip silently dropped.
+ // switching mid-run can reject transcription and drop the clip before the stopped run settles
+ assert.match(
+ host,
+ /const runBusy = useCallback\(\(\) => \{\s*if \(busyRef\.current === null\) return false;\s*toast\.info\([^)]*\);\s*return true;/,
+ );
+ // Transcribe must use the same busy guard as every other target
assert.match(
host,
- /transcribe: \(\) => \{[\s\S]*?if \(busyRef\.current !== null\) \{\s*toast\.info\([^)]*\);\s*return;\s*\}\s*if \(!transitionWorkflow\("transcribe"\)\) return;/,
+ /clipSendTargets\(clip, ttsWorkflow\)\.map\(\(id\) => \[\s*id,\s*\(\) => \{\s*if \(!runBusy\(\)\) handlers\[id\]\?\.\(\);/,
);
});