Revert back 24h (wrong repo commit)

app.py

@@ -6,7 +6,7 @@ import torch
 
 CUDA = torch.cuda.is_available()
 
-REPO_ID = "projecte-aina/tts-ca-coqui-vits-multispeaker"
+REPO_ID = "collectivat/catotron-ona"
 
 VOICE_CONVERSION_MODELS = {
     'freevc24': 'voice_conversion_models/multilingual/vctk/freevc24',
@@ -14,26 +14,17 @@ VOICE_CONVERSION_MODELS = {
     'openvoice_v2': 'voice_conversion_models/multilingual/multi-dataset/openvoice_v2',
 }
 
-my_title = "Catalan VITS Multi-Speaker TTS"
-my_description = "Model multilocutor en català basat en VITS, entrenat per Projecte AINA. Amb conversió de veu!"
-
-# Load speakers from local speakers.pth
-speakers_dict = torch.load('speakers.pth', map_location='cpu')
-SPEAKER_IDS = list(speakers_dict.keys())
-
-# Create mapping: "Speaker 1" -> actual hash ID
-SPEAKER_CHOICES = [f"Speaker {i+1}" for i in range(len(SPEAKER_IDS))]
-SPEAKER_MAP = {f"Speaker {i+1}": SPEAKER_IDS[i] for i in range(len(SPEAKER_IDS))}
+my_title = "Catotron Text-to-Speech with Voice Conversion"
+my_description = "This space allows speaker conversion on Fast Speech based 🐸 [Catotron](https://huggingface.co/collectivat/catotron-ona)." 
 
 my_examples = [
-    ["Bon dia! Com esteu avui?", "Speaker 1", True, None, 'freevc24'],
-    ["Catalunya és un país ric en cultura i tradicions.", "Speaker 1", True, None, 'freevc24'],
-    ["La intel·ligència artificial està transformant el món.", "Speaker 1", False, None, 'freevc24']
+    ["Catotron, síntesi de la parla obert i lliure en català.", True, None, 'freevc24'],
+    ["Leonor Ferrer Girabau va ser una delineant, mestra i activista barcelonina, nascuda al carrer actual de la Concòrdia del Poble-sec, que es va convertir en la primera dona a obtenir el títol de delineant a Catalunya i a l'estat.", True, None, 'freevc24'],
+    ["S'espera un dia anticiclònic amb temperatures suaus i vent fluix.", False, None, 'freevc24']
 ]
 
 my_inputs = [
-    gr.Textbox(lines=5, label="Text en català"),
-    gr.Dropdown(label="Speaker", choices=SPEAKER_CHOICES, value="Speaker 1"),
+    gr.Textbox(lines=5, label="Input Text"),
     gr.Checkbox(label="Split Sentences", 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())),
@@ -41,19 +32,19 @@ my_inputs = [
 
 my_outputs = gr.Audio(type="filepath", label="Output Audio", autoplay=True)
 
-# Download model from HF
-best_model_path = hf_hub_download(repo_id=REPO_ID, filename="model/best_model.pth")
+best_model_path = hf_hub_download(repo_id=REPO_ID, filename="fast-speech_best_model.pth") 
+config_path = hf_hub_download(repo_id=REPO_ID, filename="fast-speech_config.json")
+vocoder_model = hf_hub_download(repo_id=REPO_ID, filename="ljspeech--hifigan_v2_model_file.pth")
+vocoder_config = hf_hub_download(repo_id=REPO_ID, filename="ljspeech--hifigan_v2_config.json")
 
-api = TTS(model_path=best_model_path, config_path="config.json").to("cuda" if CUDA else "cpu")
+api = TTS(model_path=best_model_path, config_path=config_path, vocoder_path=vocoder_model, vocoder_config_path=vocoder_config).to("cuda" if CUDA else "cpu")
 
 # 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_name: str, split_sentences: bool = False, speaker_wav: str = None, voice_cv_model: str = 'freevc24'):
-    # Map "Speaker 1" to actual hash ID
-    speaker_id = SPEAKER_MAP[speaker_name]
-    
+def tts(text: str, 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(")", ",")
@@ -62,9 +53,9 @@ def tts(text: str, speaker_name: str, split_sentences: bool = False, speaker_wav
     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, speaker=speaker_id)
+            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, speaker=speaker_id)
+            api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences)
 
     return fp.name