Update handler.py

handler.py

@@ -1,4 +1,4 @@
-from typing import  Dict, Any
+from typing import  Dict, Any,Union
 import librosa
 import numpy as np
 import torch
@@ -7,6 +7,7 @@ import noisereduce as nr
 from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
 from num2tib.core import convert
 from num2tib.core import convert2text
+import base64
 import re
 import requests
 converter = pyewts.pyewts()
@@ -60,7 +61,7 @@ class EndpointHandler():
         self.vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
 
 
-    def __call__(self, data: Dict[str, Any]) -> bytes:
+    def __call__(self, data: Dict[str, Any]) -> Dict[str, Union[int, str]]:
         """_summary_
 
         Args:
@@ -86,4 +87,8 @@ class EndpointHandler():
         speaker_embedding = torch.tensor(speaker_embedding)
         speech = self.model.generate_speech(input_ids.to('cuda'), speaker_embedding.to('cuda'), vocoder=self.vocoder.to('cuda'))
         speech = nr.reduce_noise(y=speech.to('cpu'), sr=16000)
-        return speech.tobytes()
+        return {
+            "sample_rate": 16000,
+            "audio": base64.b64encode(speech.numpy()).decode("utf-8"),
+
+        }