add_zero_shot_spk

README.md

@@ -85,7 +85,6 @@ We strongly recommend that you download our pretrained `CosyVoice2-0.5B` `CosyVo
 from modelscope import snapshot_download
 snapshot_download('iic/CosyVoice2-0.5B', local_dir='pretrained_models/CosyVoice2-0.5B')
 snapshot_download('iic/CosyVoice-300M', local_dir='pretrained_models/CosyVoice-300M')
-snapshot_download('iic/CosyVoice-300M-25Hz', local_dir='pretrained_models/CosyVoice-300M-25Hz')
 snapshot_download('iic/CosyVoice-300M-SFT', local_dir='pretrained_models/CosyVoice-300M-SFT')
 snapshot_download('iic/CosyVoice-300M-Instruct', local_dir='pretrained_models/CosyVoice-300M-Instruct')
 snapshot_download('iic/CosyVoice-ttsfrd', local_dir='pretrained_models/CosyVoice-ttsfrd')
@@ -96,7 +95,6 @@ snapshot_download('iic/CosyVoice-ttsfrd', local_dir='pretrained_models/CosyVoice
 mkdir -p pretrained_models
 git clone https://www.modelscope.cn/iic/CosyVoice2-0.5B.git pretrained_models/CosyVoice2-0.5B
 git clone https://www.modelscope.cn/iic/CosyVoice-300M.git pretrained_models/CosyVoice-300M
-git clone https://www.modelscope.cn/iic/CosyVoice-300M-25Hz.git pretrained_models/CosyVoice-300M-25Hz
 git clone https://www.modelscope.cn/iic/CosyVoice-300M-SFT.git pretrained_models/CosyVoice-300M-SFT
 git clone https://www.modelscope.cn/iic/CosyVoice-300M-Instruct.git pretrained_models/CosyVoice-300M-Instruct
 git clone https://www.modelscope.cn/iic/CosyVoice-ttsfrd.git pretrained_models/CosyVoice-ttsfrd
@@ -136,6 +134,11 @@ prompt_speech_16k = load_wav('./asset/zero_shot_prompt.wav', 16000)
 for i, j in enumerate(cosyvoice.inference_zero_shot('收到好友从远方寄来的生日礼物，那份意外的惊喜与深深的祝福让我心中充满了甜蜜的快乐，笑容如花儿般绽放。', '希望你以后能够做的比我还好呦。', prompt_speech_16k, stream=False)):
     torchaudio.save('zero_shot_{}.wav'.format(i), j['tts_speech'], cosyvoice.sample_rate)
 
+# save zero_shot spk for futher usage
+assert cosyvoice.add_zero_shot_spk('希望你以后能够做的比我还好呦。', prompt_speech_16k, 'my_zero_shot_spk') is True
+for i, j in enumerate(cosyvoice.inference_zero_shot('收到好友从远方寄来的生日礼物，那份意外的惊喜与深深的祝福让我心中充满了甜蜜的快乐，笑容如花儿般绽放。', '', '', zero_shot_spk_id='my_zero_shot_spk', stream=False)):
+    torchaudio.save('zero_shot_{}.wav'.format(i), j['tts_speech'], cosyvoice.sample_rate)
+
 # fine grained control, for supported control, check cosyvoice/tokenizer/tokenizer.py#L248
 for i, j in enumerate(cosyvoice.inference_cross_lingual('在他讲述那个荒诞故事的过程中，他突然[laughter]停下来，因为他自己也被逗笑了[laughter]。', prompt_speech_16k, stream=False)):
     torchaudio.save('fine_grained_control_{}.wav'.format(i), j['tts_speech'], cosyvoice.sample_rate)
@@ -164,7 +167,7 @@ print(cosyvoice.list_available_spks())
 for i, j in enumerate(cosyvoice.inference_sft('你好，我是通义生成式语音大模型，请问有什么可以帮您的吗？', '中文女', stream=False)):
     torchaudio.save('sft_{}.wav'.format(i), j['tts_speech'], cosyvoice.sample_rate)
 
-cosyvoice = CosyVoice('pretrained_models/CosyVoice-300M') # or change to pretrained_models/CosyVoice-300M-25Hz for 25Hz inference
+cosyvoice = CosyVoice('pretrained_models/CosyVoice-300M')
 # zero_shot usage, <|zh|><|en|><|jp|><|yue|><|ko|> for Chinese/English/Japanese/Cantonese/Korean
 prompt_speech_16k = load_wav('./asset/zero_shot_prompt.wav', 16000)
 for i, j in enumerate(cosyvoice.inference_zero_shot('收到好友从远方寄来的生日礼物，那份意外的惊喜与深深的祝福让我心中充满了甜蜜的快乐，笑容如花儿般绽放。', '希望你以后能够做的比我还好呦。', prompt_speech_16k, stream=False)):

cosyvoice/cli/cosyvoice.py

@@ -66,6 +66,14 @@ class CosyVoice:
         spks = list(self.frontend.spk2info.keys())
         return spks
 
+    def add_zero_shot_spk(self, prompt_text, prompt_speech_16k, zero_shot_spk_id):
+        assert zero_shot_spk_id != '', 'do not use empty zero_shot_spk_id'
+        model_input = self.frontend.frontend_zero_shot('', prompt_text, prompt_speech_16k, self.sample_rate, '')
+        del model_input['text']
+        del model_input['text_len']
+        self.frontend.spk2info[zero_shot_spk_id] = model_input
+        return True
+
     def inference_sft(self, tts_text, spk_id, stream=False, speed=1.0, text_frontend=True):
         for i in tqdm(self.frontend.text_normalize(tts_text, split=True, text_frontend=text_frontend)):
             model_input = self.frontend.frontend_sft(i, spk_id)
@@ -77,12 +85,12 @@ class CosyVoice:
                 yield model_output
                 start_time = time.time()
 
-    def inference_zero_shot(self, tts_text, prompt_text, prompt_speech_16k, stream=False, speed=1.0, text_frontend=True):
+    def inference_zero_shot(self, tts_text, prompt_text, prompt_speech_16k, zero_shot_spk_id='', stream=False, speed=1.0, text_frontend=True):
         prompt_text = self.frontend.text_normalize(prompt_text, split=False, text_frontend=text_frontend)
         for i in tqdm(self.frontend.text_normalize(tts_text, split=True, text_frontend=text_frontend)):
             if (not isinstance(i, Generator)) and len(i) < 0.5 * len(prompt_text):
                 logging.warning('synthesis text {} too short than prompt text {}, this may lead to bad performance'.format(i, prompt_text))
-            model_input = self.frontend.frontend_zero_shot(i, prompt_text, prompt_speech_16k, self.sample_rate)
+            model_input = self.frontend.frontend_zero_shot(i, prompt_text, prompt_speech_16k, self.sample_rate, zero_shot_spk_id)
             start_time = time.time()
             logging.info('synthesis text {}'.format(i))
             for model_output in self.model.tts(**model_input, stream=stream, speed=speed):

cosyvoice/cli/frontend.py

@@ -122,7 +122,7 @@ class CosyVoiceFrontEnd:
         if isinstance(text, Generator):
             logging.info('get tts_text generator, will skip text_normalize!')
             return [text]
-        if text_frontend is False:
+        if text_frontend is False or text == '':
             return [text] if split is True else text
         text = text.strip()
         if self.use_ttsfrd:
@@ -154,24 +154,28 @@ class CosyVoiceFrontEnd:
         model_input = {'text': tts_text_token, 'text_len': tts_text_token_len, 'llm_embedding': embedding, 'flow_embedding': embedding}
         return model_input
 
-    def frontend_zero_shot(self, tts_text, prompt_text, prompt_speech_16k, resample_rate):
+    def frontend_zero_shot(self, tts_text, prompt_text, prompt_speech_16k, resample_rate, zero_shot_spk_id):
         tts_text_token, tts_text_token_len = self._extract_text_token(tts_text)
-        prompt_text_token, prompt_text_token_len = self._extract_text_token(prompt_text)
-        prompt_speech_resample = torchaudio.transforms.Resample(orig_freq=16000, new_freq=resample_rate)(prompt_speech_16k)
-        speech_feat, speech_feat_len = self._extract_speech_feat(prompt_speech_resample)
-        speech_token, speech_token_len = self._extract_speech_token(prompt_speech_16k)
-        if resample_rate == 24000:
-            # cosyvoice2, force speech_feat % speech_token = 2
-            token_len = min(int(speech_feat.shape[1] / 2), speech_token.shape[1])
-            speech_feat, speech_feat_len[:] = speech_feat[:, :2 * token_len], 2 * token_len
-            speech_token, speech_token_len[:] = speech_token[:, :token_len], token_len
-        embedding = self._extract_spk_embedding(prompt_speech_16k)
-        model_input = {'text': tts_text_token, 'text_len': tts_text_token_len,
-                       'prompt_text': prompt_text_token, 'prompt_text_len': prompt_text_token_len,
-                       'llm_prompt_speech_token': speech_token, 'llm_prompt_speech_token_len': speech_token_len,
-                       'flow_prompt_speech_token': speech_token, 'flow_prompt_speech_token_len': speech_token_len,
-                       'prompt_speech_feat': speech_feat, 'prompt_speech_feat_len': speech_feat_len,
-                       'llm_embedding': embedding, 'flow_embedding': embedding}
+        if zero_shot_spk_id == '':
+            prompt_text_token, prompt_text_token_len = self._extract_text_token(prompt_text)
+            prompt_speech_resample = torchaudio.transforms.Resample(orig_freq=16000, new_freq=resample_rate)(prompt_speech_16k)
+            speech_feat, speech_feat_len = self._extract_speech_feat(prompt_speech_resample)
+            speech_token, speech_token_len = self._extract_speech_token(prompt_speech_16k)
+            if resample_rate == 24000:
+                # cosyvoice2, force speech_feat % speech_token = 2
+                token_len = min(int(speech_feat.shape[1] / 2), speech_token.shape[1])
+                speech_feat, speech_feat_len[:] = speech_feat[:, :2 * token_len], 2 * token_len
+                speech_token, speech_token_len[:] = speech_token[:, :token_len], token_len
+            embedding = self._extract_spk_embedding(prompt_speech_16k)
+            model_input = {'prompt_text': prompt_text_token, 'prompt_text_len': prompt_text_token_len,
+                        'llm_prompt_speech_token': speech_token, 'llm_prompt_speech_token_len': speech_token_len,
+                        'flow_prompt_speech_token': speech_token, 'flow_prompt_speech_token_len': speech_token_len,
+                        'prompt_speech_feat': speech_feat, 'prompt_speech_feat_len': speech_feat_len,
+                        'llm_embedding': embedding, 'flow_embedding': embedding}
+        else:
+            model_input = self.spk2info[zero_shot_spk_id]
+        model_input['text'] = tts_text_token
+        model_input['text_len'] = tts_text_token_len
         return model_input
 
     def frontend_cross_lingual(self, tts_text, prompt_speech_16k, resample_rate):