Add OpenVoice VC models

app.py

@@ -8,6 +8,12 @@ CUDA = torch.cuda.is_available()
 
 REPO_ID = "ayymen/Coqui-TTS-Vits-shi"
 
+VOICE_CONVERSION_MODELS = {
+    'freevc24': 'voice_conversion_models/multilingual/vctk/freevc24'
+    'openvoice_v1': 'voice_conversion_models/multilingual/multi-dataset/openvoice_v1',
+    'openvoice_v2': 'voice_conversion_models/multilingual/multi-dataset/openvoice_v2',
+}
+
 my_title = "ⴰⴹⵕⵉⵚ ⵙ ⵉⵎⵙⵍⵉ - Tamazight Text-to-Speech"
 my_description = "This model is based on [VITS](https://github.com/jaywalnut310/vits), thanks to 🐸 [Coqui.ai](https://coqui.ai/)." 
 
@@ -21,6 +27,7 @@ my_examples = [
 my_inputs = [
   gr.Textbox(lines=5, label="Input Text", placeholder="The only available characters are: ⴰⴱⴳⴷⴹⴻⴼⴽⵀⵃⵄⵅⵇⵉⵊⵍⵎⵏⵓⵔⵕⵖⵙⵚⵛⵜⵟⵡⵢⵣⵥⵯ !,.:?"),
   gr.Audio(type="filepath", label="Speaker audio for voice cloning (optional)"),
+  gr.Dropdown(label="Voice Conversion Model", choices=list(VOICE_CONVERSION_MODELS.keys())),
   gr.Checkbox(label="Split Sentences (each sentence will be generated separately)", value=True)
 ]
 
@@ -31,19 +38,22 @@ config_path = hf_hub_download(repo_id=REPO_ID, filename="config.json")
 
 api = TTS(model_path=best_model_path, config_path=config_path).to("cuda" if CUDA else "cpu")
 
-# load voice conversion model
-api.load_vc_model_by_name("voice_conversion_models/multilingual/vctk/freevc24", gpu=CUDA)
+# pre-download voice conversion models
+for model in VOICE_CONVERSION_MODELS.values():
+    api.load_vc_model_by_name(model, gpu=CUDA)
 
-def tts(text: str, speaker_wav: str = None, split_sentences: bool = True):
+def tts(text: str, speaker_wav: str = None, voice_cv_model: str = 'freevc24', split_sentences: bool = True):
     # replace oov characters
     text = text.replace("\n", ". ")
     text = text.replace("(", ",")
     text = text.replace(")", ",")
+    text = text.replace('"', ",")
     text = text.replace(";", ",")
     text = text.replace("-", " ")
 
     with tempfile.NamedTemporaryFile(suffix = ".wav", delete = False) as fp:
         if speaker_wav:
+            api.load_vc_model_by_name(VOICE_CONVERSION_MODELS[voice_cv_model], gpu=CUDA)
             api.tts_with_vc_to_file(text, speaker_wav=speaker_wav, file_path=fp.name, split_sentences=split_sentences)
         else:
             api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences)

requirements.txt

@@ -1 +1 @@
-coqui-tts==0.24.3
+coqui-tts==0.25.0