diff --git a/server/voice_changer/RVC/pipeline/Pipeline.py b/server/voice_changer/RVC/pipeline/Pipeline.py
index 34612c7f..e44d0e6c 100644
--- a/server/voice_changer/RVC/pipeline/Pipeline.py
+++ b/server/voice_changer/RVC/pipeline/Pipeline.py
@@ -180,7 +180,7 @@ class Pipeline(object):
             with torch.no_grad():
                 audio1 = (
                     (
-                    torch.clip(self.inferencer.infer(feats, p_len, pitch, pitchf, sid)[0][0, 0], -1., 1.) * 32767.5 - .5
+                    torch.clip(self.inferencer.infer(feats, p_len, pitch, pitchf, sid)[0][0, 0].to(dtype=torch.float32), -1., 1.) * 32767.5 - .5
                     )
                     .data
                     .to(dtype=torch.int16)
diff --git a/server/voice_changer/RVC/pitchExtractor/CrepePitchExtractor.py b/server/voice_changer/RVC/pitchExtractor/CrepePitchExtractor.py
index d1849f02..493ef945 100644
--- a/server/voice_changer/RVC/pitchExtractor/CrepePitchExtractor.py
+++ b/server/voice_changer/RVC/pitchExtractor/CrepePitchExtractor.py
@@ -27,7 +27,7 @@ class CrepePitchExtractor(PitchExtractor):
         f0_mel_max = 1127 * np.log(1 + f0_max / 700)
 
         f0 = torchcrepe.predict(
-            torch.tensor(audio).unsqueeze(0),
+            audio.unsqueeze(0),
             sr,
             hop_length=window,
             fmin=f0_min,