From 6b73e2287c6e7520b80f853c115ea1b7a1588338 Mon Sep 17 00:00:00 2001
From: hubian <908234780@qq.com>
Date: Wed, 22 Apr 2026 18:11:13 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E6=94=B9=E7=94=A8soundfile=E4=BF=9D?=
 =?UTF-8?q?=E5=AD=98=E9=9F=B3=E9=A2=91=EF=BC=8C=E8=A7=A3=E5=86=B3torchaudi?=
 =?UTF-8?q?o=E6=A0=BC=E5=BC=8F=E9=97=AE=E9=A2=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 requirements.txt |  3 ++-
 server.py        | 12 ++++++++----
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/requirements.txt b/requirements.txt
index eaecb69..a645629 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -4,4 +4,5 @@ python-multipart==0.0.9
 torch==2.2.0
 torchaudio==2.2.0
 transformers==4.38.0
-ChatTTS==0.1.1
\ No newline at end of file
+ChatTTS
+soundfile==0.12.1
\ No newline at end of file
diff --git a/server.py b/server.py
index 4188910..a6a6925 100644
--- a/server.py
+++ b/server.py
@@ -124,11 +124,15 @@ def save_audio(audio_tensor: torch.Tensor, filename: str) -> str:
     filepath = os.path.join(AUDIO_DIR, filename)
     
     # 确保 tensor 正确形状
-    if audio_tensor.dim() == 1:
-        audio_tensor = audio_tensor.unsqueeze(0)
+    if audio_tensor.dim() == 2:
+        audio_tensor = audio_tensor.squeeze(0)
     
-    # 保存为 WAV
-    torchaudio.save(filepath, audio_tensor, SAMPLE_RATE, format="wav")
+    # 转换为 numpy
+    audio_np = audio_tensor.cpu().numpy() if audio_tensor.is_cuda else audio_tensor.numpy()
+    
+    # 使用 soundfile 保存
+    import soundfile as sf
+    sf.write(filepath, audio_np, SAMPLE_RATE)
     
     return filepath