gitea-mirror/F5-TTS
1
0
Fork 0
mirror of https://github.com/SWivid/F5-TTS.git synced 2026-07-28 13:26:59 -07:00
Code Issues Packages Projects Releases Wiki Activity

Fix #1046 tempfile related bug

Browse Source
This commit is contained in:
SWivid
2025-05-22 17:40:14 +08:00
parent f7f266cdd9
commit fb26b6d93e
2 changed files with 36 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/f5_tts/infer/infer_gradio.py
+9 -4
View File
@@ -3,6 +3,7 @@
import gc
import json
import os
import re
import tempfile
from collections import OrderedDict
@@ -189,16 +190,20 @@ def infer(
# Remove silence
if remove_silence:
with tempfile.NamedTemporaryFile(suffix=".wav") as f:
sf.write(f.name, final_wave, final_sample_rate)
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
temp_path = f.name
try:
sf.write(temp_path, final_wave, final_sample_rate)
remove_silence_for_generated_wav(f.name)
final_wave, _ = torchaudio.load(f.name)
finally:
os.unlink(temp_path)
final_wave = final_wave.squeeze().cpu().numpy()
# Save the spectrogram
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_spectrogram:
with tempfile.NamedTemporaryFile(suffix=".png", delete_on_close=False) as tmp_spectrogram:
spectrogram_path = tmp_spectrogram.name
save_spectrogram(combined_spectrogram, spectrogram_path)
save_spectrogram(combined_spectrogram, spectrogram_path)
return (final_sample_rate, final_wave), spectrogram_path, ref_text, used_seed
src/f5_tts/infer/utils_infer.py
+27 -25
View File
@@ -306,42 +306,44 @@ 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=False, suffix=".wav") as f:
aseg = AudioSegment.from_file(ref_audio_orig)
with tempfile.NamedTemporaryFile(delete_on_close=False, suffix=".wav") as f:
temp_path = f.name
# 1. try to find long silence for clipping
aseg = AudioSegment.from_file(ref_audio_orig)
# 1. try to find long silence for clipping
non_silent_segs = silence.split_on_silence(
aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=1000, seek_step=10
)
non_silent_wave = AudioSegment.silent(duration=0)
for non_silent_seg in non_silent_segs:
if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:
show_info("Audio is over 12s, clipping short. (1)")
break
non_silent_wave += non_silent_seg
# 2. try to find short silence for clipping if 1. failed
if len(non_silent_wave) > 12000:
non_silent_segs = silence.split_on_silence(
aseg, min_silence_len=1000, silence_thresh=-50, keep_silence=1000, seek_step=10
aseg, min_silence_len=100, silence_thresh=-40, keep_silence=1000, seek_step=10
)
non_silent_wave = AudioSegment.silent(duration=0)
for non_silent_seg in non_silent_segs:
if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:
show_info("Audio is over 12s, clipping short. (1)")
show_info("Audio is over 12s, clipping short. (2)")
break
non_silent_wave += non_silent_seg
# 2. try to find short silence for clipping if 1. failed
if len(non_silent_wave) > 12000:
non_silent_segs = silence.split_on_silence(
aseg, min_silence_len=100, silence_thresh=-40, keep_silence=1000, seek_step=10
)
non_silent_wave = AudioSegment.silent(duration=0)
for non_silent_seg in non_silent_segs:
if len(non_silent_wave) > 6000 and len(non_silent_wave + non_silent_seg) > 12000:
show_info("Audio is over 12s, clipping short. (2)")
break
non_silent_wave += non_silent_seg
aseg = non_silent_wave
aseg = non_silent_wave
# 3. if no proper silence found for clipping
if len(aseg) > 12000:
aseg = aseg[:12000]
show_info("Audio is over 12s, clipping short. (3)")
# 3. if no proper silence found for clipping
if len(aseg) > 12000:
aseg = aseg[:12000]
show_info("Audio is over 12s, clipping short. (3)")
aseg = remove_silence_edges(aseg) + AudioSegment.silent(duration=50)
aseg.export(f.name, format="wav")
ref_audio = f.name
aseg = remove_silence_edges(aseg) + AudioSegment.silent(duration=50)
aseg.export(temp_path, format="wav")
ref_audio = temp_path
# Cache the processed reference audio
_ref_audio_cache[audio_hash] = ref_audio