diff --git a/melo/api.py b/melo/api.py
index 236ea8f17..a3a65f574 100644
--- a/melo/api.py
+++ b/melo/api.py
@@ -122,7 +122,7 @@ def tts_to_file(self, text, speaker_id, output_path=None, sdp_ratio=0.2, noise_s
                     )[0][0, 0].data.cpu().float().numpy()
                 del x_tst, tones, lang_ids, bert, ja_bert, x_tst_lengths, speakers
                 # 
-            audio_list.append(audio)
+            audio_list.append(utils.normalize_loudness(audio))
         torch.cuda.empty_cache()
         audio = self.audio_numpy_concat(audio_list, sr=self.hps.data.sampling_rate, speed=speed)
 
diff --git a/melo/utils.py b/melo/utils.py
index bafca5a6d..2186f6c45 100644
--- a/melo/utils.py
+++ b/melo/utils.py
@@ -17,7 +17,25 @@
 
 logger = logging.getLogger(__name__)
 
-
+def normalize_loudness(audio):
+    max_val = np.max(np.abs(audio))
+    if max_val == 0:
+        return audio
+    scale_factor = (1 - 1e-6) / max_val
+    normalized_audio = audio * scale_factor
+
+    if False:
+        print(f'--- normalize ---')
+        print(f"Before normalization:")
+        print(f"  Min value: {np.min(audio)}")
+        print(f"  Max value: {np.max(audio)}")
+        print(f"  Length: {len(audio)}")
+        print(f"After normalization:")
+        print(f"  Min value: {np.min(normalized_audio)}")
+        print(f"  Max value: {np.max(normalized_audio)}")
+        print(f"  Length: {len(normalized_audio)}")
+
+    return normalized_audio
 
 def get_text_for_tts_infer(text, language_str, hps, device, symbol_to_id=None):
     norm_text, phone, tone, word2ph = clean_text(text, language_str)