Nofated095
/
so-vits-svc
mirror of https://github.com/svc-develop-team/so-vits-svc.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update voice auto-split

This commit is contained in:
ylzz1997 2023-03-20 14:08:55 +08:00
parent 849c792bcf
commit 0f0739ef14
2 changed files with 30 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
inference/infer_tool.py
View File

@ -102,6 +102,10 @@ def pad_array(arr, target_length):
pad_right = pad_width - pad_left
padded_arr = np.pad(arr, (pad_left, pad_right), 'constant', constant_values=(0, 0))
return padded_arr
def split_list_by_n(list_collection, n, pre=0):
for i in range(0, len(list_collection), n):
yield list_collection[i-pre if i-pre>=0 else i: i + n]
class Svc(object):

31
inference_main.py
View File

@ -25,6 +25,7 @@ def main():
# 一定要设置的部分
parser.add_argument('-m', '--model_path', type=str, default="logs/44k/G_0.pth", help='模型路径')
parser.add_argument('-c', '--config_path', type=str, default="configs/config.json", help='配置文件路径')
parser.add_argument('-cl', '--clip', type=float, default=0, help='音频自动切片，0为不切片，单位为秒/s')
parser.add_argument('-n', '--clean_names', type=str, nargs='+', default=["君の知らない物語-src.wav"], help='wav文件名列表，放在raw文件夹下')
parser.add_argument('-t', '--trans', type=int, nargs='+', default=[0], help='音高调整，支持正负（半音）')
parser.add_argument('-s', '--spk_list', type=str, nargs='+', default=['nen'], help='合成目标说话人名称')
@ -34,6 +35,7 @@ def main():
help='语音转换自动预测音高，转换歌声时不要打开这个会严重跑调')
parser.add_argument('-cm', '--cluster_model_path', type=str, default="logs/44k/kmeans_10000.pt", help='聚类模型路径，如果没有训练聚类则随便填')
parser.add_argument('-cr', '--cluster_infer_ratio', type=float, default=0, help='聚类方案占比，范围0-1，若没有训练聚类模型则填0即可')
parser.add_argument('-lg', '--linear_gradient', type=float, default=0, help='两端音频切片的交叉淡入长度，如果自动切片后出现人声不连贯可调整该数值，如果连贯建议采用默认值0，单位为秒/s')
# 不用动的部分
parser.add_argument('-sd', '--slice_db', type=int, default=-40, help='默认-40，嘈杂的音频可以-30，干声保留呼吸可以-50')
@ -55,6 +57,8 @@ def main():
cluster_infer_ratio = args.cluster_infer_ratio
noice_scale = args.noice_scale
pad_seconds = args.pad_seconds
clip = args.clip
lg = args.linear_gradient
infer_tool.fill_a_to_b(trans, clean_names)
for clean_name, tran in zip(clean_names, trans):
@ -65,22 +69,32 @@ def main():
wav_path = Path(raw_audio_path).with_suffix('.wav')
chunks = slicer.cut(wav_path, db_thresh=slice_db)
audio_data, audio_sr = slicer.chunks2audio(wav_path, chunks)
per_size = int(clip*audio_sr)
lg_size = int(lg*audio_sr)
lg = np.linspace(0,1,lg_size) if lg_size!=0 else 0
for spk in spk_list:
audio = []
for (slice_tag, data) in audio_data:
print(f'#=====segment start, {round(len(data) / audio_sr, 3)}s======')
length = int(np.ceil(len(data) / audio_sr * svc_model.target_sample))
if slice_tag:
print('jump empty segment')
_audio = np.zeros(length)
audio.extend(list(infer_tool.pad_array(_audio, length)))
continue
if per_size != 0:
datas = infer_tool.split_list_by_n(data, per_size,lg_size)
else:
datas = [data]
for k,dat in enumerate(datas):
per_length = int(np.ceil(len(dat) / audio_sr * svc_model.target_sample)) if clip!=0 else length
if clip!=0: print(f'###=====segment clip start, {round(len(dat) / audio_sr, 3)}s======')
# padd
pad_len = int(audio_sr * pad_seconds)
data = np.concatenate([np.zeros([pad_len]), data, np.zeros([pad_len])])
dat = np.concatenate([np.zeros([pad_len]), dat, np.zeros([pad_len])])
raw_path = io.BytesIO()
soundfile.write(raw_path, data, audio_sr, format="wav")
soundfile.write(raw_path, dat, audio_sr, format="wav")
raw_path.seek(0)
out_audio, out_sr = svc_model.infer(spk, tran, raw_path,
cluster_infer_ratio=cluster_infer_ratio,
@ -90,8 +104,15 @@ def main():
_audio = out_audio.cpu().numpy()
pad_len = int(svc_model.target_sample * pad_seconds)
_audio = _audio[pad_len:-pad_len]
audio.extend(list(infer_tool.pad_array(_audio, length)))
_audio = infer_tool.pad_array(_audio, per_length)
if lg_size!=0 and k!=0:
lg1 = audio[-lg_size:]
lg2 = _audio[0:lg_size]
lg_pre = lg1*(1-lg)+lg2*lg
audio = audio[0:-lg_size]
audio.extend(lg_pre)
_audio = _audio[lg_size:]
audio.extend(list(_audio))
key = "auto" if auto_predict_f0 else f"{tran}key"
cluster_name = "" if cluster_infer_ratio == 0 else f"_{cluster_infer_ratio}"
res_path = f'./results/{clean_name}_{key}_{spk}{cluster_name}.{wav_format}'