Sync from GitHub repo

This Space is synced from the GitHub repo: https://github.com/SWivid/F5-TTS. Please submit contributions to the Space there

app.py

@@ -42,6 +42,7 @@ from f5_tts.infer.utils_infer import (
     preprocess_ref_audio_text,
     remove_silence_for_generated_wav,
     save_spectrogram,
+    tempfile_kwargs,
 )
 from f5_tts.model import DiT, UNetT
 
@@ -190,7 +191,7 @@ def infer(
 
     # Remove silence
     if remove_silence:
-        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
+        with tempfile.NamedTemporaryFile(suffix=".wav", **tempfile_kwargs) as f:
             temp_path = f.name
         try:
             sf.write(temp_path, final_wave, final_sample_rate)
@@ -201,7 +202,7 @@ def infer(
         final_wave = final_wave.squeeze().cpu().numpy()
 
     # Save the spectrogram
-    with tempfile.NamedTemporaryFile(suffix=".png", delete_on_close=False) as tmp_spectrogram:
+    with tempfile.NamedTemporaryFile(suffix=".png", **tempfile_kwargs) as tmp_spectrogram:
         spectrogram_path = tmp_spectrogram.name
     save_spectrogram(combined_spectrogram, spectrogram_path)
 

src/f5_tts/infer/utils_infer.py

@@ -45,6 +45,8 @@ device = (
     else "cpu"
 )
 
+tempfile_kwargs = {"delete_on_close": False} if sys.version_info >= (3, 12) else {"delete": False}
+
 # -----------------------------------------
 
 target_sample_rate = 24000
@@ -306,7 +308,7 @@ def preprocess_ref_audio_text(ref_audio_orig, ref_text, show_info=print):
         ref_audio = _ref_audio_cache[audio_hash]
 
     else:  # first pass, do preprocess
-        with tempfile.NamedTemporaryFile(delete_on_close=False, suffix=".wav") as f:
+        with tempfile.NamedTemporaryFile(suffix=".wav", **tempfile_kwargs) as f:
             temp_path = f.name
 
         aseg = AudioSegment.from_file(ref_audio_orig)