diff --git a/TTS/tts/models/xtts.py b/TTS/tts/models/xtts.py
index 7cc9836a..19e3d684 100644
--- a/TTS/tts/models/xtts.py
+++ b/TTS/tts/models/xtts.py
@@ -1,5 +1,4 @@
 import os
-from contextlib import contextmanager
 from dataclasses import dataclass
 
 import librosa
@@ -8,7 +7,7 @@ import torch.nn.functional as F
 import torchaudio
 from coqpit import Coqpit
 
-from TTS.tts.layers.tortoise.audio_utils import denormalize_tacotron_mel, wav_to_univnet_mel
+from TTS.tts.layers.tortoise.audio_utils import wav_to_univnet_mel
 from TTS.tts.layers.xtts.gpt import GPT
 from TTS.tts.layers.xtts.hifigan_decoder import HifiDecoder
 from TTS.tts.layers.xtts.stream_generator import init_stream_support
@@ -71,10 +70,10 @@ def load_audio(audiopath, sampling_rate):
     # better load setting following: https://github.com/faroit/python_audio_loading_benchmark
     if audiopath[-4:] == ".mp3":
         # it uses torchaudio with sox backend to load mp3
-        audio, lsr = torchaudio.backend.sox_io_backend.load(audiopath)
+        audio, lsr = torchaudio.load(audiopath, backend="sox")
     else:
         # it uses torchaudio soundfile backend to load all the others data type
-        audio, lsr = torchaudio.backend.soundfile_backend.load(audiopath)
+        audio, lsr = torchaudio.load(audiopath, backend="soundfile")
 
     # stereo to mono if needed
     if audio.size(0) != 1: