Hecheng0625 commited on
Commit
575dc20
1 Parent(s): 31f7849

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +33 -1
README.md CHANGED
@@ -108,7 +108,39 @@ with torch.no_grad():
108
  sf.write("recon.wav", recon_wav[0][0].cpu().numpy(), 16000)
109
  ```
110
 
111
- FACodec can achieve zero-shot voice conversion with FACodecRedecoder
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
112
  ```python
113
  from Amphion.models.codec.ns3_codec import FACodecRedecoder
114
 
 
108
  sf.write("recon.wav", recon_wav[0][0].cpu().numpy(), 16000)
109
  ```
110
 
111
+ FACodec can achieve zero-shot voice conversion with FACodecEncoderV2/FACodecDecoderV2 or FACodecRedecoder
112
+ ```python
113
+ from Amphion.models.codec.ns3_codec import FACodecEncoderV2, FACodecDecoderV2
114
+
115
+ # Same parameters as FACodecEncoder/FACodecDecoder
116
+ fa_encoder_v2 = FACodecEncoderV2(...)
117
+ fa_decoder_v2 = FACodecDecoderV2(...)
118
+
119
+ encoder_v2_ckpt = hf_hub_download(repo_id="amphion/naturalspeech3_facodec", filename="ns3_facodec_encoder_v2.bin")
120
+ decoder_v2_ckpt = hf_hub_download(repo_id="amphion/naturalspeech3_facodec", filename="ns3_facodec_decoder_v2.bin")
121
+
122
+ fa_encoder_v2.load_state_dict(torch.load(encoder_v2_ckpt))
123
+ fa_decoder_v2.load_state_dict(torch.load(decoder_v2_ckpt))
124
+
125
+ with torch.no_grad():
126
+ enc_out_a = fa_encoder_v2(wav_a)
127
+ prosody_a = fa_encoder_v2.get_prosody_feature(wav_a)
128
+ enc_out_b = fa_encoder_v2(wav_b)
129
+ prosody_b = fa_encoder_v2.get_prosody_feature(wav_b)
130
+
131
+ vq_post_emb_a, vq_id_a, _, quantized, spk_embs_a = fa_decoder_v2(
132
+ enc_out_a, prosody_a, eval_vq=False, vq=True
133
+ )
134
+ vq_post_emb_b, vq_id_b, _, quantized, spk_embs_b = fa_decoder_v2(
135
+ enc_out_b, prosody_b, eval_vq=False, vq=True
136
+ )
137
+
138
+ vq_post_emb_a_to_b = fa_decoder_v2.vq2emb(vq_id_a, use_residual=False)
139
+ recon_wav_a_to_b = fa_decoder_v2.inference(vq_post_emb_a_to_b, spk_embs_b)
140
+ ```
141
+
142
+ or
143
+
144
  ```python
145
  from Amphion.models.codec.ns3_codec import FACodecRedecoder
146