try add token

app.py

@@ -32,7 +32,7 @@ from faster_whisper import WhisperModel, BatchedInferencePipeline
 from faster_whisper.vad import VadOptions
 import requests
 import base64
-from pyannote.audio import Pipeline, Inference
+from pyannote.audio import Pipeline, Inference, Model
 
 import os, sys, importlib.util, pathlib, ctypes, tempfile, wave, math
 import json
@@ -463,7 +463,9 @@ class WhisperTranscriber:
         global _embedder
         if _embedder is None:
             # window="whole" to compute one embedding per provided chunk
-            _embedder = Inference("pyannote/embedding", window="whole", device=torch.device("cuda"),use_auth_token=os.getenv("HF_TOKEN"),)
+            token = os.getenv("HF_TOKEN")
+            model = Model.from_pretrained("pyannote/embedding", use_auth_token=token)
+            _embedder = Inference(model, window="whole", device=torch.device("cuda"))
         return _embedder
 
     def assign_speakers_to_transcription(self, transcription_results, diarization_segments):