From b92361b99fd641bc5e1ff0a32e3d5058a3e526c8 Mon Sep 17 00:00:00 2001
From: Exulan <67993175+BushyToaster88@users.noreply.github.com>
Date: Fri, 3 Mar 2023 16:17:28 +1100
Subject: [PATCH] fix-torch-stft-error-on-gpus-sm-53

This pull request addresses an issue that arises when executing the finetune_speaker_v2.py script on GPUs with compute capability less than SM_53. The error occurs at line 104 of mel_processing.py, where the torch.stft() function is called with a half data type. To fix this, I updated the data type to float.
---
 mel_processing.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mel_processing.py b/mel_processing.py
index 96cbb77..8118d5c 100644
--- a/mel_processing.py
+++ b/mel_processing.py
@@ -101,8 +101,8 @@ def mel_spectrogram_torch(y, n_fft, num_mels, sampling_rate, hop_size, win_size,
     y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
     y = y.squeeze(1)
 
-    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
-                      center=center, pad_mode='reflect', normalized=False, onesided=True)
+    spec = torch.stft(y.float(), n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+        center=center, pad_mode='reflect', normalized=False, onesided=True)
 
     spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)