Support kokoro model

Dockerfile.redhat

@@ -109,6 +109,7 @@ SHELL ["/bin/bash", "-xo", "pipefail", "-c"]
 ARG JOBS=40
 ARG VERBOSE_LOGS=OFF
 ARG LTO_ENABLE=OFF
+ARG ESPEAK=1
 
 # hadolint ignore=DL3041
 RUN dnf install -y -d6 \
@@ -129,6 +130,8 @@ RUN dnf install -y -d6 \
             python3.12-pip \
             libicu-devel && \
             dnf clean all
+# Note: espeak-ng is no longer installed from dnf; it is built from source
+# by Bazel (see third_party/espeak_ng) and bundled in the release.
 
 WORKDIR /
 
@@ -234,11 +237,11 @@ RUN git clone https://github.com/$ov_tokenizers_org/openvino_tokenizers.git /ope
     fi
 
 WORKDIR /openvino_genai/
-ARG ov_genai_branch=master
-ARG ov_genai_org=openvinotoolkit
+ARG ov_genai_branch=kokoro_tts
+ARG ov_genai_repo=https://github.com/RyanMetcalfeInt8/openvino.genai.git
 # hadolint ignore=DL3003
 RUN if [ "$ov_use_binary" == "0" ]; then true ; else exit 0 ; fi ; \
-    git clone https://github.com/$ov_genai_org/openvino.genai /openvino_genai && cd /openvino_genai && git checkout $ov_genai_branch && git submodule update --init --recursive && \
+    git clone $ov_genai_repo /openvino_genai && cd /openvino_genai && git checkout $ov_genai_branch && git submodule update --init --recursive && \
     cmake -DCMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE -DCMAKE_CXX_FLAGS=" ${SDL_OPS} ${LTO_CXX_FLAGS} " -DCMAKE_SHARED_LINKER_FLAGS="${LTO_LD_FLAGS}" -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DENABLE_SYSTEM_ICU="True" -DBUILD_TOKENIZERS=OFF -DENABLE_SAMPLES=OFF -DENABLE_TOOLS=OFF -DENABLE_TESTS=OFF -DENABLE_XGRAMMAR=ON -S ./ -B ./build/ && \
     cmake --build ./build/ --parallel $JOBS && cp -vP /openvino_genai/build/openvino_genai/lib*.so* /opt/intel/openvino/runtime/lib/intel64/ && \
     cp -r /openvino_genai/src/cpp/include/* /opt/intel/openvino/runtime/include/ && \
@@ -393,6 +396,7 @@ LABEL "maintainer"="dariusz.trawinski@intel.com"
 ARG INSTALL_RPMS_FROM_URL=
 ARG INSTALL_DRIVER_VERSION="24.52.32224"
 ARG GPU=0
+ARG ESPEAK=1
 ARG debug_bazel_flags=
 LABEL bazel-build-flags=${debug_bazel_flags}
 LABEL supported-devices="CPU=1 GPU=${GPU}"
@@ -407,6 +411,8 @@ COPY ./install_redhat_gpu_drivers.sh /install_gpu_drivers.sh
 # hadolint ignore=DL3003,DL3041,SC2164,SC1091
 RUN if [ -f /usr/bin/dnf ] ; then export DNF_TOOL=dnf ; echo -e "max_parallel_downloads=8\nretries=50" >> /etc/dnf/dnf.conf ; else export DNF_TOOL=microdnf ; fi ; \
     $DNF_TOOL upgrade --setopt=install_weak_deps=0 --nodocs -y ; \
+    # espeak-ng shared library + data are shipped inside /ovms by the Bazel
+    # build when $ESPEAK == "1"; no system package is required at runtime.
     if [ "$GPU" == "1" ] ; then \
         source /install_gpu_drivers.sh && rm -rf /install_gpu_drivers.sh; \
     fi ; \
@@ -422,6 +428,11 @@ RUN if [ -f /usr/bin/dnf ] ; then export DNF_TOOL=dnf ; echo -e "max_parallel_do
 
 ENV LD_LIBRARY_PATH=/ovms/lib
 ENV PATH="$PATH:/ovms/bin"
+# When espeak-ng is built into the OVMS release, the libespeak-ng.so loaded
+# at runtime needs to locate its data tables. They are shipped under
+# /ovms/share/espeak-ng-data; point espeak-ng at that location explicitly so
+# we do not depend on any system /usr/share/espeak-ng-data.
+ENV ESPEAK_DATA_PATH=/ovms/share/espeak-ng-data
 
 COPY --from=pkg /ovms_release /ovms
 COPY --from=pkg /usr/local/lib/python3.12/site-packages/jinja2 /ovms/python_deps/jinja2

Dockerfile.ubuntu

@@ -95,6 +95,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 SHELL ["/bin/bash", "-xo", "pipefail", "-c"]
 
 ARG debug_bazel_flags="--strip=always  --config=mp_on_py_on --//:distro=ubuntu"
+ARG ESPEAK=1
 RUN if [ "$BASE_OS" == "ubuntu24" ] ; then apt-get update && \
     apt-get install -y software-properties-common --no-install-recommends; add-apt-repository 'ppa:deadsnakes/ppa' -y && \
     apt-get clean && rm -rf /var/lib/apt/lists/* ; fi
@@ -124,6 +125,8 @@ RUN apt-get update && apt-get install --no-install-recommends -y \
             vim && \
             apt-get clean && \
             rm -rf /var/lib/apt/lists/*
+# Note: espeak-ng is no longer installed from apt here; it is built from
+# source by Bazel (see third_party/espeak_ng) and bundled in the release.
 # on ubuntu 24.04 python3.12 is used as default python for ovms build and release
 # TF build needs python3.10 with numpy as it does not support python3.12
 RUN python3.10 -m pip install "numpy<2.0.0" --no-cache-dir
@@ -220,12 +223,12 @@ RUN if [ "$ov_use_binary" == "0" ]; then true ; else exit 0 ; fi ; \
     if ! [[ $debug_bazel_flags == *"_py_off"* ]]; then \
     cp build/python/* /opt/intel/openvino/python/openvino_tokenizers/ ; \
     fi
-ARG ov_genai_branch=master
-ARG ov_genai_org=openvinotoolkit
+ARG ov_genai_branch=kokoro_tts
+ARG ov_genai_repo=https://github.com/RyanMetcalfeInt8/openvino.genai.git
 WORKDIR /openvino_genai/
 # hadolint ignore=DL3003
 RUN if [ "$ov_use_binary" == "0" ]; then \
-    git clone https://github.com/$ov_genai_org/openvino.genai /openvino_genai && cd /openvino_genai && git checkout $ov_genai_branch && git submodule update --init --recursive && \
+    git clone $ov_genai_repo /openvino_genai && cd /openvino_genai && git checkout $ov_genai_branch && git submodule update --init --recursive && \
     cmake -DCMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE -DCMAKE_CXX_FLAGS=" ${SDL_OPS} " -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DENABLE_SYSTEM_ICU="True" -DBUILD_TOKENIZERS=OFF -DENABLE_SAMPLES=OFF -DENABLE_TOOLS=OFF -DENABLE_TESTS=OFF -DENABLE_XGRAMMAR=ON -S ./ -B ./build/ && \
     cmake --build ./build/ --parallel $JOBS && cp /openvino_genai/build/openvino_genai/lib*.so* /opt/intel/openvino/runtime/lib/intel64/ && \
     cp -r /openvino_genai/src/cpp/include/* /opt/intel/openvino/runtime/include/ && \
@@ -395,6 +398,7 @@ ARG INSTALL_RPMS_FROM_URL=
 ARG INSTALL_DRIVER_VERSION="24.26.30049"
 ARG GPU=0
 ARG NPU=0
+ARG ESPEAK=1
 ENV DEBIAN_FRONTEND=noninteractive
 ARG debug_bazel_flags=
 LABEL bazel-build-flags=${debug_bazel_flags}
@@ -413,6 +417,8 @@ COPY ./install_ubuntu_gpu_drivers.sh /tmp/install_gpu_drivers.sh
 # hadolint ignore=DL3003,SC2164
 RUN apt-get update ; \
     apt-get install -y --no-install-recommends curl ca-certificates libxml2 || exit 1; \
+    # espeak-ng shared library + data are shipped inside /ovms by the Bazel
+    # build when $ESPEAK == "1"; no system package is required at runtime.
     if [ "$GPU" == "1" ] ; then \
 	/tmp/install_gpu_drivers.sh ; \
     fi ; \
@@ -451,6 +457,11 @@ RUN if ! [[ $debug_bazel_flags == *"py_off"* ]]; then true ; else exit 0 ; fi ;
 
 ENV LD_LIBRARY_PATH=/ovms/lib
 ENV PATH="$PATH:/ovms/bin"
+# When espeak-ng is built into the OVMS release, the libespeak-ng.so loaded
+# at runtime needs to locate its data tables. They are shipped under
+# /ovms/share/espeak-ng-data; point espeak-ng at that location explicitly so
+# we do not depend on any system /usr/share/espeak-ng-data.
+ENV ESPEAK_DATA_PATH=/ovms/share/espeak-ng-data
 
 RUN echo "The source code of added GPL components is stored in https://storage.openvinotoolkit.org/repositories/openvino/ci_dependencies/container_gpl_sources/" > /ovms/thirdparty-licenses/GPL.txt
 USER ovms

Makefile

@@ -61,6 +61,7 @@ BUILD_TESTS ?= 0
 RUN_GPU_TESTS ?=
 GPU ?= 0
 NPU ?= 0
+ESPEAK ?= 1
 BUILD_NGINX ?= 0
 MEDIAPIPE_DISABLE ?= 0
 PYTHON_DISABLE ?= 0
@@ -146,8 +147,13 @@ else ifeq ($(findstring redhat,$(BASE_OS)),redhat)
 else
   $(error BASE_OS must be either ubuntu or redhat)
 endif
-CAPI_FLAGS = "--strip=$(STRIP)"$(BAZEL_DEBUG_BUILD_FLAGS)"  --config=mp_off_py_off"$(OV_TRACING_PARAMS)$(TARGET_DISTRO_PARAMS)
-BAZEL_DEBUG_FLAGS="--strip=$(STRIP)"$(BAZEL_DEBUG_BUILD_FLAGS)$(DISABLE_PARAMS)$(FUZZER_BUILD_PARAMS)$(OV_TRACING_PARAMS)$(TARGET_DISTRO_PARAMS)$(REPO_ENV)
+ifeq ($(ESPEAK),1)
+  ESPEAK_PARAMS = " --//:espeak=on"
+else
+  ESPEAK_PARAMS = " --//:espeak=off"
+endif
+CAPI_FLAGS = "--strip=$(STRIP)"$(BAZEL_DEBUG_BUILD_FLAGS)"  --config=mp_off_py_off"$(OV_TRACING_PARAMS)$(TARGET_DISTRO_PARAMS)$(ESPEAK_PARAMS)
+BAZEL_DEBUG_FLAGS="--strip=$(STRIP)"$(BAZEL_DEBUG_BUILD_FLAGS)$(DISABLE_PARAMS)$(FUZZER_BUILD_PARAMS)$(OV_TRACING_PARAMS)$(TARGET_DISTRO_PARAMS)$(ESPEAK_PARAMS)$(REPO_ENV)
 
 # Option to Override release image.
 # Release image OS *must have* glibc version >= glibc version on BASE_OS:
@@ -237,6 +243,7 @@ BUILD_ARGS = --build-arg http_proxy=$(HTTP_PROXY)\
 	--build-arg BASE_OS=$(BASE_OS)\
 	--build-arg INSTALL_RPMS_FROM_URL=$(INSTALL_RPMS_FROM_URL)\
 	--build-arg INSTALL_DRIVER_VERSION=$(INSTALL_DRIVER_VERSION)\
+	--build-arg ESPEAK=$(ESPEAK)\
 	--build-arg RELEASE_BASE_IMAGE=$(BASE_IMAGE_RELEASE)\
 	--build-arg JOBS=$(JOBS)\
 	--build-arg CAPI_FLAGS=$(CAPI_FLAGS)\

WORKSPACE

@@ -497,6 +497,10 @@ libgit2_engine()
 load("@ovms//third_party/drogon:drogon.bzl", "drogon_cpp")
 drogon_cpp()
 
+### espeak-ng (built from source via Bazel; gated by --//:espeak flag)
+load("@ovms//third_party/espeak_ng:espeak_ng.bzl", "espeak_ng")
+espeak_ng()
+
 # Azure Storage SDK
 new_local_repository(
     name = "azure",

common_settings.bzl

@@ -20,7 +20,7 @@
 load("@bazel_skylib//lib:selects.bzl", "selects")
 load("@mediapipe//mediapipe/framework:more_selects.bzl", "more_selects")
 load("@bazel_skylib//rules:common_settings.bzl", "string_flag")
-load("//:distro.bzl", "distro_flag")
+load("//:distro.bzl", "distro_flag", "espeak_flag")
 
 # cc_library rule wrapper that will accept the same arguments but if user will not provide
 # copts, linkopts, local_defines it will set them to the defaults
@@ -52,6 +52,7 @@ def ovms_cc_library(**kwargs):
 
 def create_config_settings():
     distro_flag()
+    espeak_flag()
     native.config_setting(
         name = "disable_mediapipe",
         define_values = {

create_package.sh

@@ -27,6 +27,27 @@ mkdir -vp /ovms_release/lib/custom_nodes
 if [ "$ov_use_binary" == "0" ] ; then cp -v /openvino_tokenizers/build/src/libopenvino_tokenizers.so /ovms_release/lib/ ; fi
 
 find /ovms/bazel-out/k8-*/bin -iname '*.so*' ! -type d ! -name "libgtest.so" ! -name "*params" ! -name "*.hana.*" ! -name "py_generate_pipeline.cpython*" !  -name "lib_node_*" ! -path "*test_python_binding*" ! -name "*libpython*" -exec cp -v {} /ovms_release/lib/ \;
+
+# Bundle espeak-ng data files when espeak was enabled in the Bazel build.
+# rules_foreign_cc places the cmake install tree under copy_<rule>/espeak-ng/
+# inside bazel-out. Both the shared library (picked up by the find above)
+# and the espeak-ng-data directory are required at runtime.
+ESPEAK_DATA_SRC=$(find /ovms/bazel-out/k8-*/bin/external/espeak_ng -type d -name 'espeak-ng-data' 2>/dev/null | head -n 1 || true)
+if [ -n "$ESPEAK_DATA_SRC" ] && [ -d "$ESPEAK_DATA_SRC" ] ; then
+    mkdir -p /ovms_release/share
+    cp -rL "$ESPEAK_DATA_SRC" /ovms_release/share/ ;
+fi
+# The version matches the espeak-ng tag pinned
+# in third_party/espeak_ng/espeak_ng.bzl (1.52.0);
+# update both places together when bumping espeak-ng.
+ESPEAK_REAL=libespeak-ng.so.1.52.0.1
+if [ -f "/ovms_release/lib/$ESPEAK_REAL" ]; then
+    cd /ovms_release/lib
+    rm -f libespeak-ng.so libespeak-ng.so.1
+    ln -s "$ESPEAK_REAL" libespeak-ng.so.1
+    ln -s "$ESPEAK_REAL" libespeak-ng.so
+    cd - >/dev/null
+fi
 if [ "$FUZZER_BUILD" == "0" ]; then mv /ovms_release/lib/libcustom_node* /ovms_release/lib/custom_nodes/; fi;
 cd /ovms_release/lib/ ; rm -f libcurl.so*
 cd /ovms_release/lib/ ; rm -f libazurestorage.so.* ; ln -s libazurestorage.so libazurestorage.so.7 ;ln -s libazurestorage.so libazurestorage.so.7.5

demos/common/export_models/export_model.py

@@ -84,9 +84,11 @@ def add_common_arguments(parser):
 parser_text2speech = subparsers.add_parser('text2speech', help='export model for text2speech endpoint')
 add_common_arguments(parser_text2speech)
 parser_text2speech.add_argument('--num_streams', default=0, type=int, help='The number of parallel execution streams to use for the models in the pipeline.', dest='num_streams')
-parser_text2speech.add_argument('--vocoder', type=str, help='The vocoder model to use for text2speech. For example microsoft/speecht5_hifigan', dest='vocoder')
-parser_text2speech.add_argument('--speaker_name', type=str, help='Name of the speaker', dest='speaker_name')
-parser_text2speech.add_argument('--speaker_path', type=str, help='Path to the speaker.bin file.', dest='speaker_path')
+parser_text2speech.add_argument('--model_type', default='speecht5', choices=['speecht5', 'kokoro'], help='Type of the source TTS model. speecht5 uses optimum-cli; kokoro uses a dedicated PyTorch->OpenVINO conversion path.', dest='model_type')
+parser_text2speech.add_argument('--vocoder', type=str, help='The vocoder model to use for speecht5. For example microsoft/speecht5_hifigan. Ignored for kokoro.', dest='vocoder')
+parser_text2speech.add_argument('--speaker_name', type=str, help='Name of the speaker (speecht5 only; for kokoro all voices from the HF repo are exported).', dest='speaker_name')
+parser_text2speech.add_argument('--speaker_path', type=str, help='Path to the speaker.bin file (speecht5 only; for kokoro all voices from the HF repo are exported).', dest='speaker_path')
+parser_text2speech.add_argument('--language', type=str, default=None, help='Default language code passed to the calculator (e.g. en-us, zh). Used mainly by kokoro.', dest='language')
 
 
 parser_speech2text = subparsers.add_parser('speech2text', help='export model for speech2text endpoint')
@@ -109,13 +111,19 @@ def add_common_arguments(parser):
       models_path: "{{model_path}}",
       plugin_config: '{ "NUM_STREAMS": "{{num_streams|default(1, true)}}" }',
       target_device: "{{target_device|default("CPU", true)}}",
-      {%- if speaker_name and speaker_path %}
+      {%- if language %}
+      language: "{{language}}",
+      {%- endif %}
+      {%- if voices %}
       voices: [
+        {%- for v in voices %}
         {
-            name: "{{speaker_name}}",
-            path: "{{speaker_path}}"
-        }
-      ]{% endif %}
+            name: "{{v.name}}",
+            path: "{{v.path}}"
+        }{% if not loop.last %},{% endif %}
+        {%- endfor %}
+      ]
+      {%- endif %}
     }
   }
 }
@@ -476,13 +484,42 @@ def export_embeddings_model_ov(model_repository_path, source_model, model_name,
     print("Created graph {}".format(os.path.join(model_repository_path, model_name, 'graph.pbtxt')))
     add_servable_to_config(config_file_path, model_name, os.path.relpath(os.path.join(model_repository_path, model_name), os.path.dirname(config_file_path)))
 
+def _list_kokoro_voices(destination_path):
+    """optimum-cli's Kokoro exporter writes per-voice speaker embeddings to
+    <destination_path>/voices/<name>.bin. Return the sorted list of voice names."""
+    voices_dir = os.path.join(destination_path, "voices")
+    if not os.path.isdir(voices_dir):
+        print("Warning: no voices/ directory found under", destination_path)
+        return []
+    return sorted(Path(p).stem for p in Path(voices_dir).glob("*.bin"))
+
 def export_text2speech_model(model_repository_path, source_model, model_name, precision, task_parameters, config_file_path):
     destination_path = os.path.join(model_repository_path, model_name)
     print("Exporting text2speech model to ",destination_path)
-    if not os.path.isdir(destination_path) or args['overwrite_models']:
-        optimum_command = "optimum-cli export openvino --model {} --weight-format {} --trust-remote-code --model-kwargs \"{{\\\"vocoder\\\": \\\"{}\\\"}}\" {}".format(source_model, precision, task_parameters['vocoder'], destination_path)
-        if os.system(optimum_command):
-            raise ValueError("Failed to export text2speech model", source_model)
+    model_type = task_parameters.get('model_type', 'speecht5')
+
+    if model_type == 'kokoro':
+        # optimum-intel registers Kokoro under library_name=kokoro / task=text-to-audio.
+        # The kokoro exporter also dumps each speaker embedding to voices/<name>.bin.
+        if not os.path.isfile(os.path.join(destination_path, 'openvino_model.xml')) or args['overwrite_models']:
+            optimum_command = "optimum-cli export openvino --model {} --task text-to-audio --weight-format {} {} --trust-remote-code {}".format(
+                source_model, precision, task_parameters['extra_quantization_params'], destination_path)
+            print('Running command:', optimum_command)
+            if os.system(optimum_command):
+                raise ValueError("Failed to export kokoro model", source_model)
+        voice_names = _list_kokoro_voices(destination_path)
+        # Render the graph with every available voice (path is relative to graph.pbtxt).
+        task_parameters['voices'] = [{'name': n, 'path': f'./voices/{n}.bin'} for n in voice_names]
+    else:
+        if not os.path.isdir(destination_path) or args['overwrite_models']:
+            if not task_parameters.get('vocoder'):
+                raise ValueError("--vocoder is required when --model_type=speecht5")
+            optimum_command = "optimum-cli export openvino --model {} --weight-format {} --trust-remote-code --model-kwargs \"{{\\\"vocoder\\\": \\\"{}\\\"}}\" {}".format(source_model, precision, task_parameters['vocoder'], destination_path)
+            if os.system(optimum_command):
+                raise ValueError("Failed to export text2speech model", source_model)
+        if task_parameters.get('speaker_name') and task_parameters.get('speaker_path'):
+            task_parameters['voices'] = [{'name': task_parameters['speaker_name'], 'path': task_parameters['speaker_path']}]
+
     gtemplate = jinja2.Environment(loader=jinja2.BaseLoader).from_string(t2s_graph_template)
     graph_content = gtemplate.render(model_path="./", **task_parameters)
     with open(os.path.join(model_repository_path, model_name, 'graph.pbtxt'), 'w') as f:

demos/common/export_models/requirements.txt

@@ -1,7 +1,7 @@
 --extra-index-url "https://download.pytorch.org/whl/cpu"
 --extra-index-url "https://storage.openvinotoolkit.org/simple/wheels/nightly"
 --extra-index-url "https://storage.openvinotoolkit.org/simple/wheels/pre-release"
-optimum-intel@git+https://github.com/huggingface/optimum-intel.git@d4dd21a3aa89c0671d85b704847ac06a378e761c
+optimum-intel@git+https://github.com/huggingface/optimum-intel.git@7a109befdfd4f9b94351135b17505efa28698210
 accelerate
 datasets
 diffusers  # for image generation
@@ -14,3 +14,8 @@ sentence_transformers==5.3.0
 sentencepiece  # Required by: transformers`
 torchvision
 requests
+# Kokoro TTS export (text2speech --model_type kokoro).
+# typer must stay <0.12 because kokoro -> misaki -> spacy imports `typer.main`,
+# which was removed in typer 0.12.
+kokoro==0.9.4
+typer<0.12

distro.bzl

@@ -42,3 +42,29 @@ def distro_flag():
         name = "not_ubuntu_build",
         negate = ":ubuntu_build",
     )
+
+# Controls whether espeak-ng is built from source (via Bazel) and bundled
+# into the OVMS release. When "off", no espeak-ng artifacts are produced
+# and the runtime will not have phonemization fallback available.
+def espeak_flag():
+    string_flag(
+        name = "espeak",
+        values = ["on", "off"],
+        build_setting_default = "on",
+    )
+    native.config_setting(
+        name = "espeak_on",
+        flag_values = {
+            "espeak": "on",
+        },
+    )
+    native.config_setting(
+        name = "espeak_off",
+        flag_values = {
+            "espeak": "off",
+        },
+    )
+    more_selects.config_setting_negation(
+        name = "not_espeak_on",
+        negate = ":espeak_on",
+    )

setupvars.bat

@@ -17,4 +17,6 @@
 set "OVMS_DIR=%~dp0"
 set "PYTHONHOME=%OVMS_DIR%\python"
 set "PATH=%OVMS_DIR%;%PYTHONHOME%;%PYTHONHOME%\Scripts;%PATH%"
+:: Point bundled espeak-ng (when packaged) at its data tables.
+if exist "%OVMS_DIR%\espeak-ng-data" set "ESPEAK_DATA_PATH=%OVMS_DIR%\espeak-ng-data"
 echo "OpenVINO Model Server Environment Initialized"

setupvars.ps1

@@ -22,4 +22,8 @@ if (Test-Path "$env:OVMS_DIR\python") {
 } else {
     $env:PATH="$env:PATH:$env:OVMS_DIR"
 }
+# Point bundled espeak-ng (when packaged) at its data tables.
+if (Test-Path "$env:OVMS_DIR\espeak-ng-data") {
+    $env:ESPEAK_DATA_PATH="$env:OVMS_DIR\espeak-ng-data"
+}
 echo "OpenVINO Model Server Environment Initialized"

src/BUILD

@@ -2287,6 +2287,12 @@ cc_binary(
             "//src/python/binding:pyovms.so",
         ],
         "//:disable_python": []
+    }) + select({
+        "//:espeak_on": [
+            "//third_party:espeak_ng",
+            "//third_party:espeak_ng_data",
+        ],
+        "//:espeak_off": [],
     }),
     # linkstatic = False,  # Use for dynamic linking when necessary
 )