update HF hub paths

Dockerfile

@@ -42,9 +42,9 @@ RUN pip install -r requirements.txt
 
 RUN huggingface-cli download BSC-LT/matcha-tts-cat-onnx matcha_multispeaker_cat_opset_15_10_steps_lastwords.onnx --local-dir $HOME/app/
 
-RUN huggingface-cli download BSC-LT/vocos-mel-22khz-onnx  mel_spec_22khz_v2.onnx --local-dir $HOME/app/
+RUN huggingface-cli download BSC-LT/vocos-mel-22khz-cat  mel_spec_22khz_cat.onnx --local-dir $HOME/app/
 
-RUN huggingface-cli download BSC-LT/vocos-mel-22khz-onnx  config.yaml --local-dir $HOME/app/
+RUN huggingface-cli download BSC-LT/vocos-mel-22khz-cat  config.yaml --local-dir $HOME/app/
 
 COPY --chown=user . $HOME/app/
 

infer_onnx.py

@@ -32,7 +32,7 @@ def process_text(i: int, text: str, device: torch.device):
 
 MODEL_PATH_MATCHA_MEL="matcha_multispeaker_cat_opset_15_10_steps_lastwords.onnx"
 MODEL_PATH_MATCHA="matcha_hifigan_multispeaker_cat.onnx"
-MODEL_PATH_VOCOS="mel_spec_22khz_v2.onnx"
+MODEL_PATH_VOCOS="mel_spec_22khz_cat.onnx"
 CONFIG_PATH="config.yaml"
 SPEAKER_ID_DICT="spk_to_id.json"
 
@@ -183,12 +183,13 @@ description = """
  
 🍵 Matcha-TTS, a new approach to non-autoregressive neural TTS, that uses conditional flow matching (similar to rectified flows) to speed up ODE-based speech synthesis
 
-For vocoders we use Hifigan universal version and Vocos trained in a catalan set of ~28 hours.
+For vocoders we use [Vocos](https://huggingface.co/BSC-LT/vocos-mel-22khz-cat) trained in a catalan set of ~28 hours.
+
+[Matcha](https://huggingface.co/BSC-LT/matcha-tts-cat-onnx) was trained using openslr69 and festcat datasets
 
-Matcha was trained using openslr69 and festcat datasets
 """
 
-article = "Training and demo by BSC."
+article = "Training and demo by The Language Technologies Unit from Barcelona Supercomputing Center."
 
 vits2_inference = gr.Interface(
     fn=tts,