Auto-download models from HF

README.md

@@ -11,7 +11,7 @@ Next-generation TTS model using flow-matching and DiT, inspired by [Stable Diffu
 
 As the first open-source TTS model that tried to combine flow-matching and DiT, **StableTTS** is a fast and lightweight TTS model for chinese, english and japanese speech generation. It has 31M parameters. 
 
-✨ **Huggingface demo:** [🤗](https://huggingface.co/spaces/KdaiP/StableTTS1.1)
+✨ **Hugging Face demo:** [🤗](https://huggingface.co/spaces/KdaiP/StableTTS1.1)
 
 ## News
 
@@ -36,7 +36,7 @@ As the first open-source TTS model that tried to combine flow-matching and DiT,
 
 ### Text-To-Mel model
 
-Download and place the model in the `./checkpoints` directory, it is ready for inference, finetuning and webui.
+By default, the inference scripts will automatically download the pretrained models from Hugging Face. For training, download the models and place them in the `checkpoints` directory.
 
 | Model Name | Task Details | Dataset | Download Link |
 |:----------:|:------------:|:-------------:|:-------------:|

api.py

@@ -16,6 +16,8 @@
 from datas.dataset import intersperse
 from utils.audio import load_and_resample_audio
 
+from cached_path import cached_path
+
 def get_vocoder(model_path, model_name='ffgan') -> nn.Module:
     if model_name == 'ffgan':
         # training or changing ffgan config is not supported in this repo
@@ -83,8 +85,8 @@ def get_params(self):
 
 if __name__ == '__main__':
     device = 'cuda' if torch.cuda.is_available() else 'cpu'
-    tts_model_path = './checkpoints/checkpoint_0.pt'
-    vocoder_model_path = './vocoders/pretrained/vocos.pt'
+    tts_model_path = str(cached_path('hf://KdaiP/StableTTS1.1/StableTTS/checkpoint_0.pt'))
+    vocoder_model_path = str(cached_path('hf://KdaiP/StableTTS1.1/vocoders/vocos.pt'))
 
     model = StableTTSAPI(tts_model_path, vocoder_model_path, 'vocos')
     model.to(device)

inference.ipynb

@@ -10,11 +10,12 @@
     "import torch\n",
     "\n",
     "from api import StableTTSAPI\n",
+    "from cached_path import cached_path\n",
     "\n",
     "device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
     "\n",
-    "tts_model_path = './checkpoints/checkpoint_0.pt' # path to StableTTS checkpoint\n",
-    "vocoder_model_path = './vocoders/pretrained/firefly-gan-base-generator.ckpt' # path to vocoder checkpoint\n",
+    "tts_model_path = str(cached_path('hf://KdaiP/StableTTS1.1/StableTTS/checkpoint_0.pt')) # or path to StableTTS checkpoint\n",
+    "vocoder_model_path = str(cached_path('hf://KdaiP/StableTTS1.1/vocoders/firefly-gan-base-generator.ckpt')) # or path to vocoder checkpoint\n",
     "vocoder_type = 'ffgan' # ffgan or vocos\n",
     "\n",
     "# vocoder_model_path = './vocoders/pretrained/vocos.pt'\n",

requirements.txt

@@ -7,6 +7,8 @@ soundfile # to make sure that torchaudio has at least one valid backend
 
 tensorboard
 
+cached_path
+
 # for monotonic_align
 numba
 

webui.py

@@ -11,10 +11,12 @@
 
 from api import StableTTSAPI
 
+from cached_path import cached_path
+
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
-tts_model_path = './checkpoints/checkpoint_0.pt'
-vocoder_model_path = './vocoders/pretrained/firefly-gan-base-generator.ckpt'
+tts_model_path = str(cached_path('hf://KdaiP/StableTTS1.1/StableTTS/checkpoint_0.pt'))
+vocoder_model_path = str(cached_path('hf://KdaiP/StableTTS1.1/vocoders/firefly-gan-base-generator.ckpt'))
 vocoder_type = 'ffgan'
 
 model = StableTTSAPI(tts_model_path, vocoder_model_path, vocoder_type).to(device)