fix(chatterbox): save audio with soundfile instead of torchaudio

containers/chatterbox/server.py

@@ -3,8 +3,8 @@ import io
 import os
 import tempfile
 
+import soundfile as sf
 import torch
-import torchaudio as ta
 from fastapi import FastAPI, File, Form, HTTPException, UploadFile
 from fastapi.responses import StreamingResponse
 from pydantic import BaseModel
@@ -79,7 +79,7 @@ def speech(req: SpeechRequest):
     )
 
     buf = io.BytesIO()
-    ta.save(buf, wav, model.sr, format="wav")
+    sf.write(buf, wav.squeeze(0).cpu().numpy(), model.sr, format="WAV")
     buf.seek(0)
     return StreamingResponse(buf, media_type="audio/wav")
 
@@ -114,6 +114,6 @@ def speech_clone(
         os.remove(tmp_path)
 
     buf = io.BytesIO()
-    ta.save(buf, wav, model.sr, format="wav")
+    sf.write(buf, wav.squeeze(0).cpu().numpy(), model.sr, format="WAV")
     buf.seek(0)
     return StreamingResponse(buf, media_type="audio/wav")