diff --git a/speech_recognition_msgs/CMakeLists.txt b/speech_recognition_msgs/CMakeLists.txt
new file mode 100644
index 0000000..3af29d0
--- /dev/null
+++ b/speech_recognition_msgs/CMakeLists.txt
@@ -0,0 +1,43 @@
+cmake_minimum_required(VERSION 3.5)
+project(speech_recognition_msgs)
+
+# Default to C++17
+if(NOT CMAKE_CXX_STANDARD)
+  set(CMAKE_CXX_STANDARD 17)
+endif()
+if(CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+  add_compile_options(-Wall -Wextra -Wpedantic)
+endif()
+
+find_package(ament_cmake REQUIRED)
+find_package(rosidl_default_generators REQUIRED)
+find_package(std_msgs REQUIRED)
+
+set(msg_files
+  msg/Grammar.msg
+  msg/PhraseRule.msg
+  msg/SpeechRecognitionCandidates.msg
+  msg/Vocabulary.msg
+)
+
+set(srv_files
+  srv/SpeechRecognition.srv
+)
+
+rosidl_generate_interfaces(${PROJECT_NAME}
+  ${msg_files}
+  ${srv_files}
+  DEPENDENCIES
+    std_msgs
+  ADD_LINTER_TESTS
+)
+
+if(BUILD_TESTING)
+  find_package(ament_lint_auto REQUIRED)
+  ament_lint_auto_find_test_dependencies()
+endif()
+
+ament_export_dependencies(rosidl_default_runtime)
+
+ament_package()
+
diff --git a/speech_recognition_msgs/msg/Grammar.msg b/speech_recognition_msgs/msg/Grammar.msg
new file mode 100644
index 0000000..f81b752
--- /dev/null
+++ b/speech_recognition_msgs/msg/Grammar.msg
@@ -0,0 +1,5 @@
+string name  # register this grammar with the name
+PhraseRule[] rules
+
+string[] categories
+Vocabulary[] vocabularies
diff --git a/speech_recognition_msgs/msg/PhraseRule.msg b/speech_recognition_msgs/msg/PhraseRule.msg
new file mode 100644
index 0000000..1da938b
--- /dev/null
+++ b/speech_recognition_msgs/msg/PhraseRule.msg
@@ -0,0 +1,2 @@
+string symbol
+string[] definition
diff --git a/speech_recognition_msgs/msg/SpeechRecognitionCandidates.msg b/speech_recognition_msgs/msg/SpeechRecognitionCandidates.msg
new file mode 100644
index 0000000..8c8dae1
--- /dev/null
+++ b/speech_recognition_msgs/msg/SpeechRecognitionCandidates.msg
@@ -0,0 +1,2 @@
+string[] transcript   # candidate words of speech-to-text API
+float32[] confidence  # confidence of transcript
\ No newline at end of file
diff --git a/speech_recognition_msgs/msg/Vocabulary.msg b/speech_recognition_msgs/msg/Vocabulary.msg
new file mode 100644
index 0000000..a2120a0
--- /dev/null
+++ b/speech_recognition_msgs/msg/Vocabulary.msg
@@ -0,0 +1,3 @@
+string name  # register vocabulary with the name.
+string[] words
+string[] phonemes
diff --git a/speech_recognition_msgs/package.xml b/speech_recognition_msgs/package.xml
new file mode 100644
index 0000000..2a88065
--- /dev/null
+++ b/speech_recognition_msgs/package.xml
@@ -0,0 +1,31 @@
+<?xml version="1.0"?>
+<?xml-model href="http://download.ros.org/schema/package_format3.xsd" schematypens="http://www.w3.org/2001/XMLSchema"?>
+<package format="3">
+  <name>speech_recognition_msgs</name>
+  <version>4.3.2</version>
+  <description>speech_recognition_msgs</description>
+  <maintainer email="furushchev@jsk.imi.i.u-tokyo.ac.jp">Yuki Furuta</maintainer>
+
+  <license>BSD</license>
+
+  <url type="website">http://ros.org/wiki/speech_recognition_msgs</url>
+
+  <author>Yuki Furuta</author>
+
+  <buildtool_depend>ament_cmake</buildtool_depend>
+
+<buildtool_depend>rosidl_default_generators</buildtool_depend>
+
+  <depend>std_msgs</depend>
+
+<exec_depend>rosidl_default_runtime</exec_depend>
+
+  <test_depend>ament_lint_auto</test_depend>
+  <test_depend>ament_lint_common</test_depend>
+  
+<member_of_group>rosidl_interface_packages</member_of_group>
+
+  <export>
+    <build_type>ament_cmake</build_type>
+  </export>
+</package>
diff --git a/speech_recognition_msgs/srv/SpeechRecognition.srv b/speech_recognition_msgs/srv/SpeechRecognition.srv
new file mode 100644
index 0000000..8dec9cf
--- /dev/null
+++ b/speech_recognition_msgs/srv/SpeechRecognition.srv
@@ -0,0 +1,9 @@
+Vocabulary vocabulary  #  Specify isolated words candidates to be recognized (Enabled on isolated word mode)
+Grammar grammar        #  Or, specify grammar to be recognized
+string grammar_name    #  Or, simply specify grammar name which is already registered on the server
+
+float32 duration
+bool quiet
+float32 threshold
+---
+speech_recognition_msgs/SpeechRecognitionCandidates result