Merge branch 'main' into main

GPT_SoVITS/inference_webui.py

@@ -324,26 +324,28 @@ def get_first(text):
     return text
 
 
-def get_cleaned_text_fianl(text,language):
+def get_cleaned_text_final(text,language):
     if language in {"en","all_zh","all_ja"}:
         phones, word2ph, norm_text = clean_text_inf(text, language)
     elif language in {"zh", "ja","auto"}:
         phones, word2ph, norm_text = nonen_clean_text_inf(text, language)
     return phones, word2ph, norm_text
 
-def get_bert_final(phones, word2ph, norm_text,language,device):
+def get_bert_final(phones, word2ph, text,language,device):
-    if text_language == "en":
+    if language == "en":
-        bert = get_bert_inf(phones, word2ph, norm_text, text_language)
+        bert = get_bert_inf(phones, word2ph, text, language)
-    elif text_language in {"zh", "ja","auto"}:
+    elif language in {"zh", "ja","auto"}:
-        bert = nonen_get_bert_inf(text, text_language)
+        bert = nonen_get_bert_inf(text, language)
-    elif text_language == "all_zh":
+    elif language == "all_zh":
-        bert = get_bert_feature(norm_text, word2ph).to(device)
+        bert = get_bert_feature(text, word2ph).to(device)
     else:
         bert = torch.zeros((1024, len(phones))).to(device)
     return bert
 
 def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切")):
     t0 = ttime()
+    prompt_language = dict_language[prompt_language]
+    text_language = dict_language[text_language]
     prompt_text = prompt_text.strip("\n")
     if (prompt_text[-1] not in splits): prompt_text += "。" if prompt_language != "en" else "."
     text = text.strip("\n")
@@ -375,10 +377,8 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         codes = vq_model.extract_latent(ssl_content)
         prompt_semantic = codes[0, 0]
     t1 = ttime()
-    prompt_language = dict_language[prompt_language]
-    text_language = dict_language[text_language]
 
-    phones1, word2ph1, norm_text1=get_cleaned_text_fianl(prompt_text, prompt_language)
+    phones1, word2ph1, norm_text1=get_cleaned_text_final(prompt_text, prompt_language)
 
     if (how_to_cut == i18n("凑四句一切")):
         text = cut1(text)
@@ -402,9 +402,8 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
             continue
         if (text[-1] not in splits): text += "。" if text_language != "en" else "."
         print(i18n("实际输入的目标文本(每句):"), text)
-        phones2, word2ph2, norm_text2 = get_cleaned_text_fianl(text, text_language)
+        phones2, word2ph2, norm_text2 = get_cleaned_text_final(text, text_language)
         bert2 = get_bert_final(phones2, word2ph2, norm_text2, text_language, device).to(dtype)
-
         bert = torch.cat([bert1, bert2], 1)
 
         all_phoneme_ids = torch.LongTensor(phones1 + phones2).to(device).unsqueeze(0)
@@ -583,7 +582,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             inp_ref = gr.Audio(label=i18n("请上传3~10秒内参考音频，超过会报错！"), type="filepath")
             prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="")
             prompt_language = gr.Dropdown(
-                label=i18n("参考音频的语种"), choices=[i18n("中文"), i18n("英文"), i18n("日文")], value=i18n("中文")
+                label=i18n("参考音频的语种"), choices=[i18n("中文"), i18n("英文"), i18n("日文"), i18n("中英混合"), i18n("日英混合"), i18n("多语种混合")], value=i18n("中文")
             )
         gr.Markdown(value=i18n("*请填写需要合成的目标文本。中英混合选中文，日英混合选日文，中日混合暂不支持，非目标语言文本自动遗弃。"))
         with gr.Row():

docs/cn/Changelog_CN.md

@@ -87,8 +87,18 @@
 
 2-引入paddlespeech的Normalizer https://github.com/RVC-Boss/GPT-SoVITS/pull/377 修复一些问题，例如：xx.xx%(带百分号类)，元/吨 会读成 元吨 而不是元每吨,下划线不再会报错
 
+### 20240207更新
+
+1-修正语种传参混乱导致中文推理效果下降 https://github.com/RVC-Boss/GPT-SoVITS/issues/391
+
+2-uvr5适配高版本librosa https://github.com/RVC-Boss/GPT-SoVITS/pull/403
+
+3-修复uvr5 inf everywhere报错的问题(is_half传参未转换bool导致恒定半精度推理，16系显卡会inf) https://github.com/RVC-Boss/GPT-SoVITS/commit/14a285109a521679f8846589c22da8f656a46ad8
+
+
 todolist：
 
 1-中文多音字推理优化
 
+2-测试集成faster whisper ASR日文英文
 

requirements.txt

@@ -5,10 +5,11 @@ librosa==0.9.2
 numba==0.56.4
 pytorch-lightning
 gradio==3.38.0
+gradio_client==0.8.1
 ffmpeg-python
 onnxruntime
 tqdm
-funasr>=1.0.0
+funasr==1.0.0
 cn2an
 pypinyin
 pyopenjtalk

tools/uvr5/lib/lib_v5/spec_utils.py

@@ -43,8 +43,8 @@ def wave_to_spectrogram(
         wave_left = np.asfortranarray(wave[0])
         wave_right = np.asfortranarray(wave[1])
 
-    spec_left = librosa.stft(wave_left, n_fft, hop_length=hop_length)
+    spec_left  = librosa.stft(wave_left,  n_fft=n_fft, hop_length=hop_length)
-    spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length)
+    spec_right = librosa.stft(wave_right, n_fft=n_fft, hop_length=hop_length)
 
     spec = np.asfortranarray([spec_left, spec_right])
 
@@ -78,7 +78,7 @@ def wave_to_spectrogram_mt(
         kwargs={"y": wave_left, "n_fft": n_fft, "hop_length": hop_length},
     )
     thread.start()
-    spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length)
+    spec_right = librosa.stft(wave_right, n_fft=n_fft, hop_length=hop_length)
     thread.join()
 
     spec = np.asfortranarray([spec_left, spec_right])
@@ -230,27 +230,31 @@ def cache_or_load(mix_path, inst_path, mp):
 
             if d == len(mp.param["band"]):  # high-end band
                 X_wave[d], _ = librosa.load(
-                    mix_path, bp["sr"], False, dtype=np.float32, res_type=bp["res_type"]
+                    mix_path, 
+                    sr       = bp["sr"],
+                    mono     = False,
+                    dtype    = np.float32,
+                    res_type = bp["res_type"]
                 )
                 y_wave[d], _ = librosa.load(
                     inst_path,
-                    bp["sr"],
+                    sr       = bp["sr"],
-                    False,
+                    mono     = False,
-                    dtype=np.float32,
+                    dtype    = np.float32,
-                    res_type=bp["res_type"],
+                    res_type = bp["res_type"],
                 )
             else:  # lower bands
                 X_wave[d] = librosa.resample(
                     X_wave[d + 1],
-                    mp.param["band"][d + 1]["sr"],
+                    orig_sr   = mp.param["band"][d + 1]["sr"],
-                    bp["sr"],
+                    target_sr = bp["sr"],
-                    res_type=bp["res_type"],
+                    res_type  = bp["res_type"],
                 )
                 y_wave[d] = librosa.resample(
                     y_wave[d + 1],
-                    mp.param["band"][d + 1]["sr"],
+                    orig_sr   = mp.param["band"][d + 1]["sr"],
-                    bp["sr"],
+                    target_sr = bp["sr"],
-                    res_type=bp["res_type"],
+                    res_type  = bp["res_type"],
                 )
 
             X_wave[d], y_wave[d] = align_wave_head_and_tail(X_wave[d], y_wave[d])
@@ -401,9 +405,9 @@ def cmb_spectrogram_to_wave(spec_m, mp, extra_bins_h=None, extra_bins=None):
                         mp.param["mid_side_b2"],
                         mp.param["reverse"],
                     ),
-                    bp["sr"],
+                    orig_sr   = bp["sr"],
-                    sr,
+                    target_sr = sr,
-                    res_type="sinc_fastest",
+                    res_type  = "sinc_fastest",
                 )
             else:  # mid
                 spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1)
@@ -418,8 +422,8 @@ def cmb_spectrogram_to_wave(spec_m, mp, extra_bins_h=None, extra_bins=None):
                         mp.param["reverse"],
                     ),
                 )
-                # wave = librosa.core.resample(wave2, bp['sr'], sr, res_type="sinc_fastest")
+                # wave = librosa.core.resample(wave2, orig_sr=bp['sr'], target_sr=sr, res_type="sinc_fastest")
-                wave = librosa.core.resample(wave2, bp["sr"], sr, res_type="scipy")
+                wave = librosa.core.resample(wave2, orig_sr=bp["sr"], target_sr=sr, res_type="scipy")
 
     return wave.T
 
@@ -506,8 +510,8 @@ def ensembling(a, specs):
 def stft(wave, nfft, hl):
     wave_left = np.asfortranarray(wave[0])
     wave_right = np.asfortranarray(wave[1])
-    spec_left = librosa.stft(wave_left, nfft, hop_length=hl)
+    spec_left = librosa.stft(wave_left, n_fft=nfft, hop_length=hl)
-    spec_right = librosa.stft(wave_right, nfft, hop_length=hl)
+    spec_right = librosa.stft(wave_right, n_fft=nfft, hop_length=hl)
     spec = np.asfortranarray([spec_left, spec_right])
 
     return spec
@@ -569,10 +573,10 @@ if __name__ == "__main__":
             if d == len(mp.param["band"]):  # high-end band
                 wave[d], _ = librosa.load(
                     args.input[i],
-                    bp["sr"],
+                    sr       = bp["sr"],
-                    False,
+                    mono     = False,
-                    dtype=np.float32,
+                    dtype    = np.float32,
-                    res_type=bp["res_type"],
+                    res_type = bp["res_type"],
                 )
 
                 if len(wave[d].shape) == 1:  # mono to stereo
@@ -580,9 +584,9 @@ if __name__ == "__main__":
             else:  # lower bands
                 wave[d] = librosa.resample(
                     wave[d + 1],
-                    mp.param["band"][d + 1]["sr"],
+                    orig_sr   = mp.param["band"][d + 1]["sr"],
-                    bp["sr"],
+                    target_sr = bp["sr"],
-                    res_type=bp["res_type"],
+                    res_type  = bp["res_type"],
                 )
 
             spec[d] = wave_to_spectrogram(

tools/uvr5/vr.py

@@ -61,19 +61,19 @@ class AudioPre:
                     _,
                 ) = librosa.core.load(  # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑
                     music_file,
-                    bp["sr"],
+                    sr       = bp["sr"],
-                    False,
+                    mono     = False,
-                    dtype=np.float32,
+                    dtype    = np.float32,
-                    res_type=bp["res_type"],
+                    res_type = bp["res_type"],
                 )
                 if X_wave[d].ndim == 1:
                     X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]])
             else:  # lower bands
                 X_wave[d] = librosa.core.resample(
                     X_wave[d + 1],
-                    self.mp.param["band"][d + 1]["sr"],
+                    orig_sr   = self.mp.param["band"][d + 1]["sr"],
-                    bp["sr"],
+                    target_sr = bp["sr"],
-                    res_type=bp["res_type"],
+                    res_type  = bp["res_type"],
                 )
             # Stft of wave source
             X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(
@@ -245,19 +245,19 @@ class AudioPreDeEcho:
                     _,
                 ) = librosa.core.load(  # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑
                     music_file,
-                    bp["sr"],
+                    sr       = bp["sr"],
-                    False,
+                    mono     = False,
-                    dtype=np.float32,
+                    dtype    = np.float32,
-                    res_type=bp["res_type"],
+                    res_type = bp["res_type"],
                 )
                 if X_wave[d].ndim == 1:
                     X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]])
             else:  # lower bands
                 X_wave[d] = librosa.core.resample(
                     X_wave[d + 1],
-                    self.mp.param["band"][d + 1]["sr"],
+                    orig_sr   = self.mp.param["band"][d + 1]["sr"],
-                    bp["sr"],
+                    target_sr = bp["sr"],
-                    res_type=bp["res_type"],
+                    res_type  = bp["res_type"],
                 )
             # Stft of wave source
             X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(

tools/uvr5/webui.py

@@ -19,7 +19,7 @@ for name in os.listdir(weight_uvr5_root):
         uvr5_names.append(name.replace(".pth", ""))
 
 device=sys.argv[1]
-is_half=sys.argv[2]
+is_half=eval(sys.argv[2])
 webui_port_uvr5=int(sys.argv[3])
 is_share=eval(sys.argv[4])