proper base64 encoding

handler.py

@@ -2,6 +2,8 @@
 from typing import Dict, Any, List
 from transformers import pipeline
 import torch
+from io import BytesIO
+import base64
 
 
 class EndpointHandler:
@@ -32,7 +34,7 @@ class EndpointHandler:
             List[Dict[str, Any]]: _description_
         """
         inputs = data["inputs"]
-        audio = inputs["audio"]  # bytes
+        audio = base64.b64decode(inputs["audio"])  # bytes
         lang = inputs["language"]
         task = inputs["task"]  # One of "translate", "transcribe"
 
@@ -41,6 +43,6 @@ class EndpointHandler:
         self.pipeline
 
         # Model inference
-        output = self.pipeline(audio.encode("latin-1"), return_timestamps="word")
+        output = self.pipeline(audio, return_timestamps="word")
 
         return output

test_handler.py

@@ -1,13 +1,16 @@
 """Test the custom handler."""
 from handler import EndpointHandler
 import os
+import base64
 
 handler = EndpointHandler(path="whisper-tiny")
 
 with open("frank.wav", "rb") as file:
     audio = file.read()
 
-payload = {"audio": audio, "language": "nl", "task": "transcribe"}
+audio = base64.b64encode(audio)
+
+payload = {"inputs": {"audio": audio, "language": "nl", "task": "transcribe"}}
 
 output = handler(payload)