feat: Refine TTS retry logic to handle client-side transport failures and enhance voice management within the TTS context.

src/app/api/tts/route.ts

@@ -13,6 +13,9 @@ import { isAuthEnabled } from '@/lib/server/auth/config';
 import { getClientIp } from '@/lib/server/rate-limit/request-ip';
 import { getOrCreateDeviceId, setDeviceIdCookie } from '@/lib/server/rate-limit/device-id';
 
+export const runtime = 'nodejs';
+export const maxDuration = 60;
+
 function attachDeviceIdCookie(response: NextResponse, deviceId: string | null, didCreate: boolean) {
   if (didCreate && deviceId) {
     setDeviceIdCookie(response, deviceId);
@@ -237,6 +240,10 @@ export async function POST(req: NextRequest) {
     const openai = new OpenAI({
       apiKey: openApiKey,
       baseURL: openApiBaseUrl,
+      // Keep retry policy centralized in this route's fetchTTSBufferWithRetry.
+      maxRetries: 0,
+      // Keep upstream timeout below route max duration so we can return a controlled error.
+      timeout: Number(process.env.TTS_UPSTREAM_TIMEOUT_MS ?? 45_000),
     });
 
     const normalizedVoice = (

src/components/player/VoicesControl.tsx

@@ -1,6 +1,7 @@
 'use client';
 
 import { useConfig } from '@/contexts/ConfigContext';
+import { useTTS } from '@/contexts/TTSContext';
 import { useCallback } from 'react';
 import { VoicesControlBase } from '@/components/player/VoicesControlBase';
 
@@ -8,7 +9,8 @@ export const VoicesControl = ({ availableVoices, setVoiceAndRestart }: {
   availableVoices: string[];
   setVoiceAndRestart: (voice: string) => void;
 }) => {
-  const { voice, ttsModel, ttsProvider } = useConfig();
+  const { ttsModel, ttsProvider } = useConfig();
+  const { voice } = useTTS();
   const onChangeVoice = useCallback((nextVoice: string) => setVoiceAndRestart(nextVoice), [setVoiceAndRestart]);
 
   return (

src/contexts/EPUBContext.tsx

@@ -457,11 +457,12 @@ export function EPUBProvider({ children }: { children: ReactNode }) {
             instructions: effectiveModel === 'gpt-4o-mini-tts' ? ttsInstructions : undefined
           };
 
+          // Allow one narrow client retry for transient browser->/api/tts transport failures.
+          // HTTP failures are not retried client-side.
           const retryOptions: TTSRetryOptions = {
             maxRetries: 2,
-            initialDelay: 5000,
-            maxDelay: 10000,
-            backoffFactor: 2
+            initialDelay: 300,
+            maxDelay: 300,
           };
 
           const audioBuffer = await withRetry(
@@ -623,11 +624,12 @@ export function EPUBProvider({ children }: { children: ReactNode }) {
         instructions: effectiveModel === 'gpt-4o-mini-tts' ? ttsInstructions : undefined
       };
 
+      // Allow one narrow client retry for transient browser->/api/tts transport failures.
+      // HTTP failures are not retried client-side.
       const retryOptions: TTSRetryOptions = {
         maxRetries: 2,
-        initialDelay: 5000,
-        maxDelay: 10000,
-        backoffFactor: 2
+        initialDelay: 300,
+        maxDelay: 300,
       };
 
       const audioBuffer = await withRetry(

src/contexts/PDFContext.tsx

@@ -513,11 +513,12 @@ export function PDFProvider({ children }: { children: ReactNode }) {
           instructions: effectiveModel === 'gpt-4o-mini-tts' ? ttsInstructions : undefined
         };
 
+        // Allow one narrow client retry for transient browser->/api/tts transport failures.
+        // HTTP failures are not retried client-side.
         const retryOptions: TTSRetryOptions = {
-          maxRetries: 3,
-          initialDelay: 1000,
-          maxDelay: 5000,
-          backoffFactor: 2
+          maxRetries: 2,
+          initialDelay: 300,
+          maxDelay: 300,
         };
 
         try {
@@ -686,11 +687,12 @@ export function PDFProvider({ children }: { children: ReactNode }) {
         instructions: effectiveModel === 'gpt-4o-mini-tts' ? ttsInstructions : undefined
       };
 
+      // Allow one narrow client retry for transient browser->/api/tts transport failures.
+      // HTTP failures are not retried client-side.
       const retryOptions: TTSRetryOptions = {
-        maxRetries: 3,
-        initialDelay: 1000,
-        maxDelay: 5000,
-        backoffFactor: 2
+        maxRetries: 2,
+        initialDelay: 300,
+        maxDelay: 300,
       };
 
       const audioBuffer: TTSAudioBuffer = await withRetry(

src/contexts/TTSContext.tsx

@@ -67,6 +67,7 @@ declare global {
  */
 interface TTSContextType extends TTSPlaybackState {
   // Voice settings
+  voice: string;
   availableVoices: string[];
 
   // Alignment metadata for the current sentence
@@ -905,23 +906,33 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement
     if (availableVoices.length > 0) {
       // Allow Kokoro multi-voice strings (e.g., "voice1(0.5)+voice2(0.5)") for any provider
       const isKokoro = isKokoroModel(configTTSModel);
+      const fallbackVoice = configVoice || availableVoices[0];
 
       if (isKokoro) {
         // If Kokoro and we have any voice string (including plus/weights), don't override it.
-        // Only default when voice is empty.
+        // Only default when local voice is empty.
         if (!voice) {
-          setVoice(availableVoices[0]);
+          setVoice(fallbackVoice);
         }
         return;
       }
 
-      if (!voice || !availableVoices.includes(voice)) {
+      // For non-Kokoro, only force a fallback when there is no active local voice.
+      // If a persisted config voice exists, keep it rather than overriding from a
+      // potentially stale in-flight voices response during reload.
+      if (!voice) {
+        console.log(`Voice "${voice || '(empty)'}" not set. Using "${fallbackVoice}"`);
+        setVoice(fallbackVoice);
+        return;
+      }
+
+      if (!configVoice && !availableVoices.includes(voice)) {
         console.log(`Voice "${voice || '(empty)'}" not found in available voices. Using "${availableVoices[0]}"`);
         setVoice(availableVoices[0]);
         // Don't save to config - just use it temporarily until user explicitly selects one
       }
     }
-  }, [availableVoices, voice, configTTSModel]);
+  }, [availableVoices, voice, configVoice, configTTSModel]);
 
   /**
    * Generates and plays audio for the current sentence
@@ -1029,11 +1040,12 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement
         instructions: ttsModel === 'gpt-4o-mini-tts' ? ttsInstructions : undefined,
       };
 
+      // Allow one narrow client retry for transient browser->/api/tts transport failures.
+      // HTTP failures are not retried client-side.
       const retryOptions: TTSRetryOptions = {
-        maxRetries: 3,
-        initialDelay: 1000,
-        maxDelay: 5000,
-        backoffFactor: 2
+        maxRetries: 2,
+        initialDelay: 300,
+        maxDelay: 300,
       };
 
       onTTSStart();
@@ -1727,6 +1739,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement
     currDocPage,
     currDocPageNumber,
     currDocPages,
+    voice,
     availableVoices,
     togglePlay,
     skipForward,
@@ -1751,6 +1764,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement
     currDocPage,
     currDocPageNumber,
     currDocPages,
+    voice,
     availableVoices,
     togglePlay,
     skipForward,

src/hooks/audio/useVoiceManagement.ts

@@ -1,6 +1,6 @@
 'use client';
 
-import { useState, useCallback } from 'react';
+import { useState, useCallback, useRef } from 'react';
 import { getVoices } from '@/lib/client/api/audiobooks';
 
 const DEFAULT_VOICES = ['alloy', 'ash', 'coral', 'echo', 'fable', 'onyx', 'nova', 'sage', 'shimmer'];
@@ -20,8 +20,10 @@ export function useVoiceManagement(
   ttsModel: string | undefined
 ) {
   const [availableVoices, setAvailableVoices] = useState<string[]>([]);
+  const fetchSeqRef = useRef(0);
 
   const fetchVoices = useCallback(async () => {
+    const fetchSeq = ++fetchSeqRef.current;
     try {
       console.log('Fetching voices...');
       const data = await getVoices({
@@ -31,10 +33,13 @@ export function useVoiceManagement(
         'x-tts-model': ttsModel || 'tts-1',
         'Content-Type': 'application/json',
       });
-      
+
+      // Ignore stale responses from older provider/model fetches.
+      if (fetchSeq !== fetchSeqRef.current) return;
       setAvailableVoices(data.voices || DEFAULT_VOICES);
     } catch (error) {
       console.error('Error fetching voices:', error);
+      if (fetchSeq !== fetchSeqRef.current) return;
       // Set available voices to default openai voices
       setAvailableVoices(DEFAULT_VOICES);
     }

src/lib/client/api/audiobooks.ts

@@ -62,6 +62,26 @@ export const withRetry = async <T>(
         break;
       }
 
+      // Narrow client retries to transport-level failures only.
+      // If we got an HTTP status from /api/tts, do not retry from the client.
+      // Server-side /api/tts already applies upstream retry logic.
+      const message = lastError.message.toLowerCase();
+      const isTransportFailure =
+        status === undefined &&
+        (
+          lastError instanceof TypeError ||
+          message.includes('failed to fetch') ||
+          message.includes('networkerror') ||
+          message.includes('network request failed') ||
+          message.includes('load failed') ||
+          message.includes('fetch failed') ||
+          message.includes('econnreset') ||
+          message.includes('etimedout')
+        );
+      if (!isTransportFailure) {
+        break;
+      }
+
       if (attempt === maxRetries - 1) {
         break;
       }