fix

app.py

@@ -11,6 +11,9 @@ import numpy as np
 import soundfile
 import random
 from audio2numpy import open_audio
+from spleeter.separator import Separator
+from spleeter.audio.adapter import AudioAdapter
+from pydub import AudioSegment
 
 import logging
 
@@ -78,6 +81,8 @@ model = SynthesizerInfer(
 load_svc_model("vits_pretrain/sovits5.0-48k-debug.pth", model)
 model.eval()
 model.to(device)
+separator = Separator('spleeter:2stems')
+audio_loader = AudioAdapter.default()
 
 
 def svc_change(argswave, argsspk):
@@ -167,17 +172,34 @@ def svc_main(sid, input_audio):
     if input_audio is None:
         return "You need to upload an audio", None
     sampling_rate, audio = input_audio
-    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
-    if len(audio.shape) > 1:
-        audio = librosa.to_mono(audio.transpose(1, 0))
+    #
+    prediction = separator.separate(audio)
+    vocals, accompaniment = prediction["vocals"], prediction["accompaniment"]
+
+    # 歌词和伴奏是否有问题
+    audio_loader.save('vocals_111.wav', vocals, sampling_rate)
+    audio_loader.save('accompaniment_111.wav', accompaniment, sampling_rate)
+
+    vocals = (vocals / np.iinfo(vocals.dtype).max).astype(np.float32)
+    if len(vocals.shape) > 1:
+        vocals = librosa.to_mono(vocals.transpose(1, 0))
     if sampling_rate != 16000:
-        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
-    if len(audio) > 16000 * 100:
-        audio = audio[:16000 * 100]
+        vocals = librosa.resample(vocals, orig_sr=sampling_rate, target_sr=16000)
+    if len(vocals) > 16000 * 100:
+        vocals = vocals[:16000 * 100]
     wav_path = "temp.wav"
-    soundfile.write(wav_path, audio, 16000, format="wav")
-    out_audio = svc_change(wav_path, f"configs/singers/singer00{sid}.npy")
-    return "Success", (48000, out_audio)
+    soundfile.write(wav_path, vocals, 16000, format="wav")
+
+    out_vocals = svc_change(wav_path, f"configs/singers/singer00{sid}.npy")
+
+    audio_loader.save('out_vocals_111.wav', out_vocals, 48000)
+
+    sound1 = AudioSegment.from_file("out_vocals_111.wav")
+    sound2 = AudioSegment.from_file("accompaniment_111.wav")
+
+    played_togther = sound1.overlay(sound2)
+
+    return "Success", (48000, played_togther)
 
 
 def auto_search(name):

music/search.py

@@ -87,5 +87,5 @@ def get_random_spit(songinfo):
     if right_idx > length:
         right_idx = length
     middle_30s = song[left_idx:right_idx]
-    middle_30s.export(save_path, format="mp3")
+    middle_30s.export(save_path, format="wav")
     return save_path

requirements.txt

@@ -14,4 +14,5 @@ tqdm
 librosa
 pydub
 musicdl
-audio2numpy
+audio2numpy
+spleeter