feat: enhance TTS generation and voice resolution with support for custom models and improved audio format detection

Author: liudonghua123

lib/audio/tts-providers.ts

@@ -172,12 +172,21 @@ export async function generateTTS(
 
 /**
  * OpenAI TTS implementation (direct API call with explicit UTF-8 encoding)
+ * Supports both official OpenAI TTS and OpenAI-compatible TTS APIs
  */
 async function generateOpenAITTS(
   config: TTSModelConfig,
   text: string,
 ): Promise<TTSGenerationResult> {
-  const baseUrl = config.baseUrl || TTS_PROVIDERS['openai-tts'].defaultBaseUrl;
+  // Determine baseUrl based on provider type
+  let baseUrl: string;
+  
+  if (config.baseUrl) {
+    baseUrl = config.baseUrl;
+  } else {
+    // For official OpenAI TTS, use default baseUrl
+    baseUrl = TTS_PROVIDERS['openai-tts'].defaultBaseUrl || 'https://api.openai.com/v1';
+  }
 
   // Use gpt-4o-mini-tts for best quality and intelligent realtime applications
   const response = await fetch(`${baseUrl}/audio/speech`, {
@@ -200,9 +209,30 @@ async function generateOpenAITTS(
   }
 
   const arrayBuffer = await response.arrayBuffer();
+  
+  // Extract audio format from Content-Type header
+  const contentType = response.headers.get('content-type') || 'audio/mpeg';
+  let format = 'mp3'; // default
+  
+  if (contentType.includes('audio/wav')) {
+    format = 'wav';
+  } else if (contentType.includes('audio/mpeg') || contentType.includes('audio/mp3')) {
+    format = 'mp3';
+  } else if (contentType.includes('audio/ogg')) {
+    format = 'ogg';
+  } else if (contentType.includes('audio/aac')) {
+    format = 'aac';
+  } else if (contentType.includes('audio/flac')) {
+    format = 'flac';
+  } else if (contentType.includes('audio/webm')) {
+    format = 'webm';
+  } else if (contentType.includes('audio/opus')) {
+    format = 'opus';
+  }
+
   return {
     audio: new Uint8Array(arrayBuffer),
-    format: 'mp3',
+    format,
   };
 }
 

lib/audio/voice-resolver.ts

@@ -78,11 +78,17 @@ export interface ProviderWithVoices {
  * Get all available providers and their voices for the voice picker UI.
  * A provider is available if it has an API key or is server-configured.
  * Browser-native-tts is excluded (no static voice list).
+ * Includes both built-in models and custom models for OpenAI-compatible providers.
  */
 export function getAvailableProvidersWithVoices(
   ttsProvidersConfig: Record<
     string,
-    { apiKey?: string; enabled?: boolean; isServerConfigured?: boolean }
+    {
+      apiKey?: string;
+      enabled?: boolean;
+      isServerConfigured?: boolean;
+      customModels?: Array<{ id: string; name: string }>;
+    }
   >,
 ): ProviderWithVoices[] {
   const result: ProviderWithVoices[] = [];
@@ -101,6 +107,8 @@ export function getAvailableProvidersWithVoices(
 
       // Build model groups
       const modelGroups: ModelVoiceGroup[] = [];
+      
+      // Add built-in models
       if (config.models.length > 0) {
         for (const model of config.models) {
           const compatibleVoices = config.voices
@@ -112,8 +120,23 @@ export function getAvailableProvidersWithVoices(
             voices: compatibleVoices,
           });
         }
-      } else {
-        // Provider has no model concept (Azure, Browser Native, Doubao)
+      }
+      
+      // Add custom models for OpenAI-compatible TTS
+      const customModels = providerConfig?.customModels || [];
+      if (providerId === 'openai-compatible-tts' && customModels.length > 0) {
+        for (const model of customModels) {
+          // All voices are compatible with custom models
+          modelGroups.push({
+            modelId: model.id,
+            modelName: model.name,
+            voices: allVoices,
+          });
+        }
+      }
+      
+      // If no models at all, add default group (for providers without model concept)
+      if (modelGroups.length === 0) {
         modelGroups.push({
           modelId: '',
           modelName: config.name,