feat: chunked TTS generation for long text (engine-agnostic)

app/src/components/ServerSettings/ConnectionForm.tsx

@@ -4,6 +4,7 @@ import { useForm } from 'react-hook-form';
 import * as z from 'zod';
 import { Button } from '@/components/ui/button';
 import { Card, CardContent, CardHeader, CardTitle } from '@/components/ui/card';
+import { Checkbox } from '@/components/ui/checkbox';
 import {
   Form,
   FormControl,
@@ -14,10 +15,10 @@ import {
   FormMessage,
 } from '@/components/ui/form';
 import { Input } from '@/components/ui/input';
-import { Checkbox } from '@/components/ui/checkbox';
+import { Slider } from '@/components/ui/slider';
 import { useToast } from '@/components/ui/use-toast';
-import { useServerStore } from '@/stores/serverStore';
 import { usePlatform } from '@/platform/PlatformContext';
+import { useServerStore } from '@/stores/serverStore';
 
 const connectionSchema = z.object({
   serverUrl: z.string().url('Please enter a valid URL'),
@@ -33,6 +34,8 @@ export function ConnectionForm() {
   const setKeepServerRunningOnClose = useServerStore((state) => state.setKeepServerRunningOnClose);
   const mode = useServerStore((state) => state.mode);
   const setMode = useServerStore((state) => state.setMode);
+  const maxChunkChars = useServerStore((state) => state.maxChunkChars);
+  const setMaxChunkChars = useServerStore((state) => state.setMaxChunkChars);
   const { toast } = useToast();
 
   const form = useForm<ConnectionFormValues>({
@@ -59,11 +62,7 @@ export function ConnectionForm() {
   }
 
   return (
-    <Card
-      role="region"
-      aria-label="Server Connection"
-      tabIndex={0}
-    >
+    <Card role="region" aria-label="Server Connection" tabIndex={0}>
       <CardHeader>
         <CardTitle>Server Connection</CardTitle>
       </CardHeader>
@@ -153,6 +152,32 @@ export function ConnectionForm() {
             </div>
           </div>
         )}
+
+        <div className="mt-6 pt-6 border-t">
+          <div className="space-y-3">
+            <div className="flex items-center justify-between">
+              <label htmlFor="maxChunkChars" className="text-sm font-medium leading-none">
+                Auto-chunking limit
+              </label>
+              <span className="text-sm tabular-nums text-muted-foreground">
+                {maxChunkChars} chars
+              </span>
+            </div>
+            <Slider
+              id="maxChunkChars"
+              value={[maxChunkChars]}
+              onValueChange={([value]) => setMaxChunkChars(value)}
+              min={100}
+              max={2000}
+              step={50}
+              aria-label="Auto-chunking character limit"
+            />
+            <p className="text-sm text-muted-foreground">
+              Long text is split into chunks at sentence boundaries before generating. Lower values
+              can improve quality for long outputs. Default is 800.
+            </p>
+          </div>
+        </div>
       </CardContent>
     </Card>
   );

app/src/lib/api/types.ts

@@ -36,6 +36,7 @@ export interface GenerationRequest {
   model_size?: '1.7B' | '0.6B';
   engine?: 'qwen' | 'luxtts' | 'chatterbox' | 'chatterbox_turbo';
   instruct?: string;
+  max_chunk_chars?: number;
 }
 
 export interface GenerationResponse {

app/src/lib/hooks/useGenerationForm.ts

@@ -9,9 +9,10 @@ import { useGeneration } from '@/lib/hooks/useGeneration';
 import { useModelDownloadToast } from '@/lib/hooks/useModelDownloadToast';
 import { useGenerationStore } from '@/stores/generationStore';
 import { usePlayerStore } from '@/stores/playerStore';
+import { useServerStore } from '@/stores/serverStore';
 
 const generationSchema = z.object({
-  text: z.string().min(1, 'Text is required').max(5000),
+  text: z.string().min(1, 'Text is required').max(50000),
   language: z.enum(LANGUAGE_CODES as [LanguageCode, ...LanguageCode[]]),
   seed: z.number().int().optional(),
   modelSize: z.enum(['1.7B', '0.6B']).optional(),
@@ -31,6 +32,7 @@ export function useGenerationForm(options: UseGenerationFormOptions = {}) {
   const generation = useGeneration();
   const setAudioWithAutoPlay = usePlayerStore((state) => state.setAudioWithAutoPlay);
   const setIsGenerating = useGenerationStore((state) => state.setIsGenerating);
+  const maxChunkChars = useServerStore((state) => state.maxChunkChars);
   const [downloadingModelName, setDownloadingModelName] = useState<string | null>(null);
   const [downloadingDisplayName, setDownloadingDisplayName] = useState<string | null>(null);
 
@@ -110,6 +112,7 @@ export function useGenerationForm(options: UseGenerationFormOptions = {}) {
         model_size: isQwen ? data.modelSize : undefined,
         engine,
         instruct: isQwen ? data.instruct || undefined : undefined,
+        max_chunk_chars: maxChunkChars,
       });
 
       toast({

app/src/stores/serverStore.ts

@@ -13,6 +13,9 @@ interface ServerStore {
 
   keepServerRunningOnClose: boolean;
   setKeepServerRunningOnClose: (keepRunning: boolean) => void;
+
+  maxChunkChars: number;
+  setMaxChunkChars: (value: number) => void;
 }
 
 export const useServerStore = create<ServerStore>()(
@@ -29,6 +32,9 @@ export const useServerStore = create<ServerStore>()(
 
       keepServerRunningOnClose: false,
       setKeepServerRunningOnClose: (keepRunning) => set({ keepServerRunningOnClose: keepRunning }),
+
+      maxChunkChars: 800,
+      setMaxChunkChars: (value) => set({ maxChunkChars: value }),
     }),
     {
       name: 'voicebox-server',

backend/main.py

@@ -824,18 +824,24 @@ async def download_chatterbox_turbo_background():
             engine=engine,
         )
 
-        audio, sample_rate = await tts_model.generate(
-            data.text,
-            voice_prompt,
-            data.language,
-            data.seed,
-            data.instruct,
-        )
+        from .utils.chunked_tts import generate_chunked
 
-        # Trim trailing silence/hallucination for Chatterbox output
+        # Resolve per-chunk trim function for engines that need it
+        trim_fn = None
         if engine in ("chatterbox", "chatterbox_turbo"):
             from .utils.audio import trim_tts_output
-            audio = trim_tts_output(audio, sample_rate)
+            trim_fn = trim_tts_output
+
+        audio, sample_rate = await generate_chunked(
+            tts_model,
+            data.text,
+            voice_prompt,
+            language=data.language,
+            seed=data.seed,
+            instruct=data.instruct,
+            max_chunk_chars=data.max_chunk_chars,
+            trim_fn=trim_fn,
+        )
 
         # Calculate duration
         duration = len(audio) / sample_rate
@@ -949,18 +955,23 @@ async def stream_speech(
         data.profile_id, db, engine=engine,
     )
 
-    audio, sample_rate = await tts_model.generate(
-        data.text,
-        voice_prompt,
-        data.language,
-        data.seed,
-        data.instruct,
-    )
+    from .utils.chunked_tts import generate_chunked
 
-    # Trim trailing silence/hallucination for Chatterbox output
+    trim_fn = None
     if engine in ("chatterbox", "chatterbox_turbo"):
         from .utils.audio import trim_tts_output
-        audio = trim_tts_output(audio, sample_rate)
+        trim_fn = trim_tts_output
+
+    audio, sample_rate = await generate_chunked(
+        tts_model,
+        data.text,
+        voice_prompt,
+        language=data.language,
+        seed=data.seed,
+        instruct=data.instruct,
+        max_chunk_chars=data.max_chunk_chars,
+        trim_fn=trim_fn,
+    )
 
     wav_bytes = tts.audio_to_wav_bytes(audio, sample_rate)
 

backend/models.py

@@ -52,12 +52,13 @@ class Config:
 class GenerationRequest(BaseModel):
     """Request model for voice generation."""
     profile_id: str
-    text: str = Field(..., min_length=1, max_length=5000)
+    text: str = Field(..., min_length=1, max_length=50000)
     language: str = Field(default="en", pattern="^(zh|en|ja|ko|de|fr|ru|pt|es|it|he)$")
     seed: Optional[int] = Field(None, ge=0)
     model_size: Optional[str] = Field(default="1.7B", pattern="^(1\\.7B|0\\.6B)$")
     instruct: Optional[str] = Field(None, max_length=500)
     engine: Optional[str] = Field(default="qwen", pattern="^(qwen|luxtts|chatterbox|chatterbox_turbo)$")
+    max_chunk_chars: int = Field(default=800, ge=100, le=5000, description="Max characters per chunk for long text splitting")
 
 
 class GenerationResponse(BaseModel):