Use multi-speaker model

app.py

@@ -6,7 +6,7 @@ import torch
 
 CUDA = torch.cuda.is_available()
 
-REPO_ID = "ayymen/Coqui-TTS-Vits-shi"
+REPO_ID = "ayymen/Coqui-TTS-Vits-Multispeaker"
 
 VOICE_CONVERSION_MODELS = {
     'freevc24': 'voice_conversion_models/multilingual/vctk/freevc24',
@@ -14,26 +14,36 @@ VOICE_CONVERSION_MODELS = {
     'openvoice_v2': 'voice_conversion_models/multilingual/multi-dataset/openvoice_v2',
 }
 
+VARIANTS = {"Tachelhit": "shi", "Tarifit": "rif"}
+
+SPEAKERS = ["yan", "sin", "idj"]
+
 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/)." 
 
 my_examples = [
-  ["ⴰⵣⵓⵍ. ⵎⴰⵏⵣⴰⴽⵉⵏ?"],
-  ["ⵡⴰ ⵜⴰⵎⵖⴰⵔⵜ ⵎⴰ ⴷ ⵓⴽⴰⵏ ⵜⵙⴽⵔⵜ?"],
-  ["ⴳⵏ ⴰⴷ ⴰⴽ ⵉⵙⵙⴳⵏ ⵕⴱⴱⵉ ⵉⵜⵜⵓ ⴽ."],
-  ["ⴰⵔⵔⴰⵡ ⵏ ⵍⵀⵎⵎ ⵢⵓⴽⵔ ⴰⵖ ⵉⵀⴷⵓⵎⵏ ⵏⵏⵖ!"]
+    ["ⴰⵣⵓⵍ. ⵎⴰⵏⵣⴰⴽⵉⵏ?", "shi", "yan", True],
+    ["ⵡⴰ ⵜⴰⵎⵖⴰⵔⵜ ⵎⴰ ⴷ ⵓⴽⴰⵏ ⵜⵙⴽⵔⵜ?", "shi", "sin", False],
+    ["ⴳⵏ ⴰⴷ ⴰⴽ ⵉⵙⵙⴳⵏ ⵕⴱⴱⵉ ⵉⵜⵜⵓ ⴽ.", "shi", "yan", False],
+    ["ⴰⵔⵔⴰⵡ ⵏ ⵍⵀⵎⵎ ⵢⵓⴽⵔ ⴰⵖ ⵉⵀⴷⵓⵎⵏ ⵏⵏⵖ!", "shi", "yan", False],
+    ["ⴰⵣⵓⵍ. ⵎⴰⵎⵛ ⵜⴷⵊⵉⵜ?", "rif", "idj", True],
+    ["ⴰⵇⵎⵎⵓⵎ ⵉⵇⵏⴻⵏ ⵓⵔ ⵜ ⵜⵜⵉⴷⴼⵏ ⵉⵣⴰⵏ.", "rif", "idj", False],
+    ["ⵇⵇⵉⵎ ⵅ ⵜⴰⴷⴷⴰⵔⵜ ⵏⵏⵛ!", "rif", "idj", False],
+    ["ⵜⴻⵜⵜⵏ ⴰⴳ ⵡⵓⵛⵛⵏ, ⵜⵜⵔⵓⵏ ⵅ ⵓⵎⴽⵙⴰ.", "rif", "idj", False]
 ]
 
 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)
+    gr.Textbox(lines=5, label="Input Text", placeholder="The only available characters are: ⴰⴱⴳⴷⴹⴻⴼⴽⵀⵃⵄⵅⵇⵉⵊⵍⵎⵏⵓⵔⵕⵖⵙⵚⵛⵜⵟⵡⵢⵣⵥⵯ !,.:?"),
+    gr.Dropdown(label="Variant", choices=list(VARIANTS.items()), value="shi"),
+    gr.Dropdown(label="Speaker", choices=SPEAKERS, value="yan"),
+    gr.Checkbox(label="Split Sentences (each sentence will be generated separately)", value=False),
+    gr.Audio(type="filepath", label="Speaker audio for voice cloning (optional)"),
+    gr.Dropdown(label="Voice Conversion Model", choices=list(VOICE_CONVERSION_MODELS.keys())),
 ]
 
 my_outputs = gr.Audio(type="filepath", label="Output Audio", autoplay=True)
 
-best_model_path = hf_hub_download(repo_id=REPO_ID, filename="best_model.pth") 
+best_model_path = hf_hub_download(repo_id=REPO_ID, filename="checkpoint_390000.pth") 
 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")
@@ -42,21 +52,22 @@ api = TTS(model_path=best_model_path, config_path=config_path).to("cuda" if CUDA
 for model in VOICE_CONVERSION_MODELS.values():
     api.load_vc_model_by_name(model, gpu=CUDA)
 
-def tts(text: str, speaker_wav: str = None, voice_cv_model: str = 'freevc24', split_sentences: bool = True):
+def tts(text: str, variant: str = "shi", speaker: str = "yan", split_sentences: bool = False, speaker_wav: str = None, voice_cv_model: str = 'freevc24'):
     # replace oov characters
     text = text.replace("\n", ". ")
     text = text.replace("(", ",")
     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)
+            api.tts_with_vc_to_file(text, speaker_wav=speaker_wav, file_path=fp.name, split_sentences=split_sentences, speaker=speaker, language=variant)
         else:
-            api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences)
+            api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences, speaker=speaker, language=variant)
 
     return fp.name 
 

language_ids.json

@@ -0,0 +1,4 @@
+{
+    "rif": 0,
+    "shi": 1
+}

speakers.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a9df430489a8bf3eac98f38325dbdbd8d986fa731787724406062bacac5a471
+size 864