Implement 2-step audio-to-audio process

backend/app/Services/GeminiService.php

@@ -218,20 +218,19 @@ class GeminiService
      */
     public static function generateAudioResponse(string $apiKey, string $systemPrompt, string $userMessage, string $voiceName = 'Puck'): ?array
     {
-        $url = 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent?key=' . $apiKey;
+        $url = 'https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-flash-tts-preview:generateContent?key=' . $apiKey;
+
+        $parts = [];
+        if (!empty($systemPrompt)) {
+            $parts[] = ['text' => "System instruction: " . $systemPrompt];
+        }
+        $parts[] = ['text' => $userMessage];
 
         $payload = json_encode([
             'contents' => [
                 [
                     'role' => 'user',
-                    'parts' => [
-                        ['text' => $userMessage]
-                    ]
-                ]
-            ],
-            'systemInstruction' => [
-                'parts' => [
-                    ['text' => $systemPrompt]
+                    'parts' => $parts
                 ]
             ],
             'generationConfig' => [
@@ -280,72 +279,14 @@ class GeminiService
      */
     public static function generateAudioResponseFromAudio(string $apiKey, string $systemPrompt, string $audioBase64, string $mimeType, string $voiceName = 'Puck'): ?array
     {
-        $url = 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent?key=' . $apiKey;
-
-        if (strpos($mimeType, ';') !== false) {
-            $mimeType = trim(explode(';', $mimeType)[0]);
-        }
-
-        $payload = json_encode([
-            'contents' => [
-                [
-                    'role' => 'user',
-                    'parts' => [
-                        [
-                            'inlineData' => [
-                                'mimeType' => $mimeType,
-                                'data' => $audioBase64
-                            ]
-                        ],
-                        [
-                            'text' => "استمع إلى التسجيل الصوتي المرفق وأجب عليه مباشرة بصوتك بناءً على الإرشادات المحددة."
-                        ]
-                    ]
-                ]
-            ],
-            'systemInstruction' => [
-                'parts' => [
-                    ['text' => $systemPrompt]
-                ]
-            ],
-            'generationConfig' => [
-                'responseModalities' => ['AUDIO'],
-                'speechConfig' => [
-                    'voiceConfig' => [
-                        'prebuiltVoiceConfig' => [
-                            'voiceName' => $voiceName
-                        ]
-                    ]
-                ]
-            ]
-        ]);
-
-        $ch = curl_init($url);
-        curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
-        curl_setopt($ch, CURLOPT_POST, true);
-        curl_setopt($ch, CURLOPT_POSTFIELDS, $payload);
-        curl_setopt($ch, CURLOPT_HTTPHEADER, [
-            'Content-Type: application/json'
-        ]);
-        curl_setopt($ch, CURLOPT_TIMEOUT, 45); // 45 seconds timeout for audio-to-audio generation
-
-        $response = curl_exec($ch);
-        $httpCode = curl_getinfo($ch, CURLINFO_HTTP_CODE);
-        curl_close($ch);
-
-        if ($httpCode !== 200) {
-            error_log("[Gemini Audio-to-Audio Response Error] HTTP " . $httpCode . " | Response: " . $response);
+        // Step 1: Use gemini-flash-lite-latest (which supports audio input) to understand the audio message and generate a text reply
+        $replyText = self::generateResponseFromAudio($apiKey, $systemPrompt, $audioBase64, $mimeType);
+        if (empty($replyText)) {
+            error_log("[Gemini Audio-to-Audio Error] Could not generate text response from audio.");
             return null;
         }
 
-        $data = json_decode($response, true);
-        $part = $data['candidates'][0]['content']['parts'][0] ?? null;
-        if ($part && isset($part['inlineData'])) {
-            return [
-                'audio' => $part['inlineData']['data'],
-                'mimeType' => $part['inlineData']['mimeType'] ?? 'audio/mp4'
-            ];
-        }
-        return null;
+        // Step 2: Use gemini-3.1-flash-tts-preview to convert the text response into a native audio voice note
+        return self::generateAudioResponse($apiKey, $systemPrompt, $replyText, $voiceName);
     }
 }

backend/public/test_audio.php

@@ -135,71 +135,19 @@ if ($successfulModel && $audioResponse && !empty($audioResponse['audio'])) {
     echo "ℹ️ [Gemini] Audio MimeType: " . $audioResponse['mimeType'] . "\n";
     echo "ℹ️ [Gemini] Audio Size: " . strlen($audioResponse['audio']) . " base64 chars\n";
     
-    // 3. Test Audio-to-Audio conversion using the successful model
-    echo "\n--- Testing Audio-to-Audio (Speech-to-Speech) ---\n";
+    // 3. Test Audio-to-Audio conversion using the Service
+    echo "\n--- Testing Audio-to-Audio (Speech-to-Speech) via GeminiService ---\n";
     $startTime = microtime(true);
     
-    // We temporarily override the model inside GeminiService for this test, but since GeminiService is not updated yet,
-    // we'll run a direct curl call for testing audio-to-audio:
-    $url = 'https://generativelanguage.googleapis.com/v1beta/models/' . $successfulModel . ':generateContent?key=' . $apiKey;
-    $payload2 = json_encode([
-        'contents' => [
-            [
-                'role' => 'user',
-                'parts' => [
-                    [
-                        'inlineData' => [
-                            'mimeType' => $audioResponse['mimeType'],
-                            'data' => $audioResponse['audio']
-                        ]
-                    ],
-                    [
-                        'text' => "استمع إلى التسجيل الصوتي المرفق وأجب عليه مباشرة بصوتك بناءً على الإرشادات المحددة."
-                    ]
-                ]
-            ]
-        ],
-        'systemInstruction' => [
-            'parts' => [['text' => $systemPrompt]]
-        ],
-        'generationConfig' => [
-            'responseModalities' => ['AUDIO'],
-            'speechConfig' => [
-                'voiceConfig' => [
-                    'prebuiltVoiceConfig' => [
-                        'voiceName' => 'Puck'
-                    ]
-                ]
-            ]
-        ]
-    ]);
-    
-    $ch = curl_init($url);
-    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
-    curl_setopt($ch, CURLOPT_POST, true);
-    curl_setopt($ch, CURLOPT_POSTFIELDS, $payload2);
-    curl_setopt($ch, CURLOPT_HTTPHEADER, ['Content-Type: application/json']);
-    curl_setopt($ch, CURLOPT_TIMEOUT, 30);
-    $response2 = curl_exec($ch);
-    $httpCode2 = curl_getinfo($ch, CURLINFO_HTTP_CODE);
-    curl_close($ch);
-    
+    $audioResponse2 = GeminiService::generateAudioResponseFromAudio($apiKey, $systemPrompt, $audioResponse['audio'], $audioResponse['mimeType'], 'Puck');
     $elapsedTime2 = round(microtime(true) - $startTime, 2);
     
-    if ($httpCode2 === 200) {
-        $data2 = json_decode($response2, true);
-        $part2 = $data2['candidates'][0]['content']['parts'][0] ?? null;
-        if ($part2 && isset($part2['inlineData'])) {
-            echo "✅ [Gemini] Successfully generated Audio-to-Audio response in {$elapsedTime2} seconds!\n";
-            echo "ℹ️ [Gemini] Audio MimeType: " . ($part2['inlineData']['mimeType'] ?? 'audio/mp4') . "\n";
-            echo "ℹ️ [Gemini] Audio Size: " . strlen($part2['inlineData']['data']) . " base64 chars\n";
-        } else {
-            echo "❌ [Gemini] Audio-to-Audio response did not contain inlineData.\n";
-        }
+    if ($audioResponse2 && !empty($audioResponse2['audio'])) {
+        echo "✅ [Gemini] Successfully generated Audio-to-Audio response in {$elapsedTime2} seconds!\n";
+        echo "ℹ️ [Gemini] Audio MimeType: " . $audioResponse2['mimeType'] . "\n";
+        echo "ℹ️ [Gemini] Audio Size: " . strlen($audioResponse2['audio']) . " base64 chars\n";
     } else {
-        $errorData2 = json_decode($response2, true);
-        $errMsg2 = $errorData2['error']['message'] ?? 'Unknown error';
-        echo "❌ [Gemini] Audio-to-Audio generation failed. HTTP {$httpCode2}: {$errMsg2}\n";
+        echo "❌ [Gemini] Audio-to-Audio generation failed.\n";
     }
 } else {
     echo "\n❌ [Gemini] All model trials for audio response generation failed.\n";