ENH: Update ChatTTS

.github/workflows/python.yaml

@@ -152,11 +152,10 @@ jobs:
             ${{ env.SELF_HOST_PYTHON }} -m conda install -c conda-forge pynini=2.1.5
             ${{ env.SELF_HOST_PYTHON }} -m pip install -U nemo_text_processing
             ${{ env.SELF_HOST_PYTHON }} -m pip install -U omegaconf~=2.3.0
-            ${{ env.SELF_HOST_PYTHON }} -m pip install -U vector_quantize_pytorch
-            ${{ env.SELF_HOST_PYTHON }} -m pip install -U vocos
             ${{ env.SELF_HOST_PYTHON }} -m pip install -U WeTextProcessing
             ${{ env.SELF_HOST_PYTHON }} -m pip install -U librosa
             ${{ env.SELF_HOST_PYTHON }} -m pip install -U xxhash
+            ${{ env.SELF_HOST_PYTHON }} -m pip install -U "ChatTTS>0.1"
             ${{ env.SELF_HOST_PYTHON }} -m pytest --timeout=1500 \
               -W ignore::PendingDeprecationWarning \
               --cov-config=setup.cfg --cov-report=xml --cov=xinference xinference/model/image/tests/test_stable_diffusion.py

setup.cfg

@@ -109,12 +109,11 @@ all =
     torchvision  # For deepseek VL
     FlagEmbedding  # For rerank
     omegaconf~=2.3.0  # For ChatTTS
-    vector_quantize_pytorch  # For ChatTTS
-    vocos  # For ChatTTS
     nemo_text_processing  # For ChatTTS
     WeTextProcessing  # For ChatTTS
     librosa  # For ChatTTS
     xxhash  # For ChatTTS
+    ChatTTS>0.1
     boto3>=1.28.55,<1.28.65 # For tensorizer
     tensorizer~=2.9.0
 intel =
@@ -155,12 +154,11 @@ image =
     controlnet_aux
 audio =
     omegaconf~=2.3.0
-    vector_quantize_pytorch
-    vocos
     nemo_text_processing
     WeTextProcessing
     librosa
     xxhash
+    ChatTTS>0.1
 doc =
     ipython>=6.5.0
     sphinx>=3.0.0

xinference/model/audio/chattts.py

@@ -38,21 +38,19 @@ def __init__(
         self._kwargs = kwargs
 
     def load(self):
+        import ChatTTS
         import torch
 
-        from xinference.thirdparty import ChatTTS
-
         torch._dynamo.config.cache_size_limit = 64
         torch._dynamo.config.suppress_errors = True
         torch.set_float32_matmul_precision("high")
         self._model = ChatTTS.Chat()
-        self._model.load_models(
-            source="local", local_path=self._model_path, compile=True
-        )
+        self._model.load(source="custom", custom_path=self._model_path, compile=True)
 
     def speech(
         self, input: str, voice: str, response_format: str = "mp3", speed: float = 1.0
     ):
+        import ChatTTS
         import numpy as np
         import torch
         import torchaudio
@@ -71,7 +69,9 @@ def speech(
 
         default = 5
         infer_speed = int(default * speed)
-        params_infer_code = {"spk_emb": rnd_spk_emb, "prompt": f"[speed_{infer_speed}]"}
+        params_infer_code = ChatTTS.Chat.InferCodeParams(
+            prompt=f"[speed_{infer_speed}]", spk_emb=rnd_spk_emb
+        )
 
         assert self._model is not None
         wavs = self._model.infer([input], params_infer_code=params_infer_code)

xinference/model/audio/tests/test_chattts.py

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
+import os.path
 import tempfile
 
 
@@ -40,5 +40,6 @@ def test_chattts(setup):
     response = client.audio.speech.create(
         model=model_uid, input=input_string, voice="echo"
     )
-    with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as f:
+    with tempfile.NamedTemporaryFile(suffix=".mp3", delete=True) as f:
         response.stream_to_file(f.name)
+        assert os.stat(f.name).st_size > 0