deepgeek/CosyVoice
2
0
Fork 0
mirror of https://github.com/FunAudioLLM/CosyVoice.git synced 2026-02-05 18:09:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

keep only embedding mean as spk embedding

Browse Source
This commit is contained in:
lyuxiang.lx
2024-07-10 00:21:56 +08:00
parent ee9e87b4d3
commit 6a3e44242a
2 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
cosyvoice/dataset/processor.py
View File

@@ -167,7 +167,7 @@ def parse_embedding(data, normalize, mode='train'):
""" """
for sample in data: for sample in data:
sample['utt_embedding'] = torch.tensor(sample['utt_embedding'], dtype=torch.float32) sample['utt_embedding'] = torch.tensor(sample['utt_embedding'], dtype=torch.float32)
sample['spk_embedding'] = torch.stack([torch.tensor(i, dtype=torch.float32) for i in sample['spk_embedding']], dim=0).mean(dim=0) sample['spk_embedding'] = torch.tensor(sample['spk_embedding'], dtype=torch.float32)
if normalize: if normalize:
sample['utt_embedding'] = F.normalize(sample['utt_embedding'], dim=0) sample['utt_embedding'] = F.normalize(sample['utt_embedding'], dim=0)
sample['spk_embedding'] = F.normalize(sample['spk_embedding'], dim=0) sample['spk_embedding'] = F.normalize(sample['spk_embedding'], dim=0)

2
tools/extract_embedding.py
View File

@@ -53,6 +53,8 @@ def main(args):
if spk not in spk2embedding: if spk not in spk2embedding:
spk2embedding[spk] = [] spk2embedding[spk] = []
spk2embedding[spk].append(embedding) spk2embedding[spk].append(embedding)
for k, v in spk2embedding.items():
spk2embedding[k] = torch.tensor(v).mean(dim=0, keepdim=True).tolist()
torch.save(utt2embedding, '{}/utt2embedding.pt'.format(args.dir)) torch.save(utt2embedding, '{}/utt2embedding.pt'.format(args.dir))
torch.save(spk2embedding, '{}/spk2embedding.pt'.format(args.dir)) torch.save(spk2embedding, '{}/spk2embedding.pt'.format(args.dir))