Spaces:
Running
on
Zero
Running
on
Zero
| # Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu) | |
| # | |
| # Licensed under the Apache License, Version 2.0 (the "License"); | |
| # you may not use this file except in compliance with the License. | |
| # You may obtain a copy of the License at | |
| # | |
| # http://www.apache.org/licenses/LICENSE-2.0 | |
| # | |
| # Unless required by applicable law or agreed to in writing, software | |
| # distributed under the License is distributed on an "AS IS" BASIS, | |
| # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
| # See the License for the specific language governing permissions and | |
| # limitations under the License. | |
| import torch | |
| class CosyVoiceModel: | |
| def __init__(self, | |
| llm: torch.nn.Module, | |
| flow: torch.nn.Module, | |
| hift: torch.nn.Module): | |
| self.device = torch.device('cuda') | |
| self.llm = llm | |
| self.flow = flow | |
| self.hift = hift | |
| def load(self, llm_model, flow_model, hift_model): | |
| self.llm.load_state_dict(torch.load(llm_model, map_location='cpu')) | |
| self.llm.to(self.device).eval() | |
| self.flow.load_state_dict(torch.load(flow_model, map_location='cpu')) | |
| self.flow.to(self.device).eval() | |
| self.hift.load_state_dict(torch.load(hift_model, map_location='cpu')) | |
| self.hift.to(self.device).eval() | |
| def inference(self, text, text_len, flow_embedding, llm_embedding=torch.zeros(0, 192), | |
| prompt_text=torch.zeros(1, 0, dtype=torch.int32), prompt_text_len=torch.zeros(1, dtype=torch.int32), | |
| llm_prompt_speech_token=torch.zeros(1, 0, dtype=torch.int32), llm_prompt_speech_token_len=torch.zeros(1, dtype=torch.int32), | |
| flow_prompt_speech_token=torch.zeros(1, 0, dtype=torch.int32), flow_prompt_speech_token_len=torch.zeros(1, dtype=torch.int32), | |
| prompt_speech_feat=torch.zeros(1, 0, 80), prompt_speech_feat_len=torch.zeros(1, dtype=torch.int32)): | |
| tts_speech_token = self.llm.inference(text=text.to(self.device), | |
| text_len=text_len.to(self.device), | |
| prompt_text=prompt_text.to(self.device), | |
| prompt_text_len=prompt_text_len.to(self.device), | |
| prompt_speech_token=llm_prompt_speech_token.to(self.device), | |
| prompt_speech_token_len=llm_prompt_speech_token_len.to(self.device), | |
| embedding=llm_embedding.to(self.device), | |
| beam_size=1, | |
| sampling=25, | |
| max_token_text_ratio=30, | |
| min_token_text_ratio=3) | |
| tts_mel = self.flow.inference(token=tts_speech_token, | |
| token_len=torch.tensor([tts_speech_token.size(1)], dtype=torch.int32).to(self.device), | |
| prompt_token=flow_prompt_speech_token.to(self.device), | |
| prompt_token_len=flow_prompt_speech_token_len.to(self.device), | |
| prompt_feat=prompt_speech_feat.to(self.device), | |
| prompt_feat_len=prompt_speech_feat_len.to(self.device), | |
| embedding=flow_embedding.to(self.device)) | |
| tts_speech = self.hift.inference(mel=tts_mel).cpu() | |
| return {'tts_speech': tts_speech} | |