diff --git a/xinference/deploy/docker/Dockerfile b/xinference/deploy/docker/Dockerfile
index 1975adb5eb..7ede36a9f0 100644
--- a/xinference/deploy/docker/Dockerfile
+++ b/xinference/deploy/docker/Dockerfile
@@ -26,6 +26,7 @@ ENV LD_LIBRARY_PATH $LD_LIBRARY_PATH:/usr/local/lib/python3.10/dist-packages/nvi
 ARG PIP_INDEX=https://pypi.org/simple
 RUN pip install --upgrade -i "$PIP_INDEX" pip && \
     pip install -i "$PIP_INDEX" "diskcache>=5.6.1" "jinja2>=2.11.3" && \
+    pip install -i "$PIP_INDEX" matcha-tts && \
     # use pre-built whl package for llama-cpp-python, otherwise may core dump when init llama in some envs
     pip install "llama-cpp-python>=0.2.82" -i https://abetlen.github.io/llama-cpp-python/whl/cu124 && \
     pip install -i "$PIP_INDEX" --upgrade-strategy only-if-needed -r /opt/inference/xinference/deploy/docker/requirements.txt && \
diff --git a/xinference/deploy/docker/cpu.Dockerfile b/xinference/deploy/docker/cpu.Dockerfile
index d7bd45c463..a22eac50d4 100644
--- a/xinference/deploy/docker/cpu.Dockerfile
+++ b/xinference/deploy/docker/cpu.Dockerfile
@@ -19,6 +19,7 @@ ENV PATH $NVM_DIR/versions/node/v$NODE_VERSION/bin:$PATH
 
 ARG PIP_INDEX=https://pypi.org/simple
 RUN python -m pip install --upgrade -i "$PIP_INDEX" pip && \
+    pip install -i "$PIP_INDEX" matcha-tts && \
     pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu && \
     pip install -i "$PIP_INDEX" --upgrade-strategy only-if-needed -r /opt/inference/xinference/deploy/docker/requirements_cpu.txt && \
     CMAKE_ARGS="-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python && \
diff --git a/xinference/deploy/docker/requirements.txt b/xinference/deploy/docker/requirements.txt
index 5408194367..7ee4776a71 100644
--- a/xinference/deploy/docker/requirements.txt
+++ b/xinference/deploy/docker/requirements.txt
@@ -52,7 +52,7 @@ torchaudio  # For ChatTTS
 ChatTTS>0.1
 xxhash  # For ChatTTS
 HyperPyYAML  # For CosyVoice
-matcha-tts>=0.0.7  # For CosyVoice
+# matcha-tts>=0.0.7  # For CosyVoice
 onnxruntime-gpu==1.16.0; sys_platform == 'linux'  # For CosyVoice
 onnxruntime==1.16.0; sys_platform == 'darwin' or sys_platform == 'windows'  # For CosyVoice
 openai-whisper  # For CosyVoice
diff --git a/xinference/deploy/docker/requirements_cpu.txt b/xinference/deploy/docker/requirements_cpu.txt
index 00a33dae9c..48d6a74e8d 100644
--- a/xinference/deploy/docker/requirements_cpu.txt
+++ b/xinference/deploy/docker/requirements_cpu.txt
@@ -49,7 +49,7 @@ torchaudio  # For ChatTTS
 ChatTTS>0.1
 xxhash  # For ChatTTS
 HyperPyYAML  # For CosyVoice
-matcha-tts>=0.0.7  # For CosyVoice
+# matcha-tts>=0.0.7  # For CosyVoice
 onnxruntime-gpu==1.16.0; sys_platform == 'linux'  # For CosyVoice
 onnxruntime==1.16.0; sys_platform == 'darwin' or sys_platform == 'windows'  # For CosyVoice
 openai-whisper  # For CosyVoice