samples: tflite-micro: add micro_speech sample with OpenAMP on i.MX8MP

This sample demonstrates TensorFlow Lite Micro speech recognition
capabilities on Zephyr RTOS, specifically designed for i.MX8MP platforms
with OpenAMP support for inter-core communication.

The sample:
- Processes audio input at 16 kHz, S16_LE, mono format.
- Uses resource table for shared memory management between cores.

Board tested: imx8mp_evk_mimx8ml8_adsp

Known limitations:
- If two commands are spoken within 1000 ms,
the second command may not be detected.
- If a command lasts longer than 1000 ms,
it may be detected as two separate commands.

Link to Linux application:
https://github.com/thong-phn/gsoc2025-linux-app

Source of micro_speech model:
https://github.com/tensorflow/tflite-micro/tree/main/tensorflow/lite/micro/examples/micro_speech

Signed-off-by: Thong Phan <[email protected]>

Author: Thong Phan

samples/modules/tflite-micro/micro_speech/CMakeLists.txt

@@ -0,0 +1,70 @@
+cmake_minimum_required(VERSION 3.20.0)
+
+find_package(Zephyr REQUIRED HINTS $ENV{ZEPHYR_BASE})
+project(micro_speech_openamp)
+
+# These samples use local static initialization. Since Zephyr doesn't support the
+# C++ ABI for thread-safe initialization of local statics and the constructors don't
+# appear to require thread safety, we turn it off in the C++ compiler.
+set(NO_THREADSAFE_STATICS $<TARGET_PROPERTY:compiler-cpp,no_threadsafe_statics>)
+zephyr_compile_options($<$<COMPILE_LANGUAGE:CXX>:${NO_THREADSAFE_STATICS}>)
+
+# Define common paths
+set(OPTIONAL_TFLITE_DIR $ENV{ZEPHYR_BASE}/../optional/modules/lib/tflite-micro/signal)
+
+file(
+        GLOB
+        app_sources
+        src/*
+        src/inference/*
+)
+
+target_include_directories(
+        app
+        PRIVATE
+        src
+        ${ZEPHYR_TFLITE_MICRO_MODULE_DIR}/signal
+        ${ZEPHYR_TFLITE_MICRO_MODULE_DIR}/signal/micro/kernels/
+        ${ZEPHYR_TFLITE_MICRO_MODULE_DIR}/signal/src
+        ${LIBMETAL_INCLUDE_DIR}
+        ${OPENAMP_INCLUDE_DIR}
+        ${PLATFORM_DIR}
+)
+
+target_sources(
+	app
+	PRIVATE
+	${app_sources}
+
+	# Signal processing
+	${OPTIONAL_TFLITE_DIR}/src/fft_auto_scale.cc
+	${OPTIONAL_TFLITE_DIR}/src/max_abs.cc
+	${OPTIONAL_TFLITE_DIR}/src/msb_32.cc
+	${OPTIONAL_TFLITE_DIR}/src/energy.cc
+	${OPTIONAL_TFLITE_DIR}/src/filter_bank.cc
+	${OPTIONAL_TFLITE_DIR}/src/pcan_argc_fixed.cc
+	${OPTIONAL_TFLITE_DIR}/src/filter_bank_log.cc
+	${OPTIONAL_TFLITE_DIR}/src/log.cc
+	${OPTIONAL_TFLITE_DIR}/src/filter_bank_spectral_subtraction.cc
+
+	# Micro kernels
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/energy.cc
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/fft_auto_scale_common.cc
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/filter_bank_square_root_common.cc
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/filter_bank_spectral_subtraction.cc
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/filter_bank.cc
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/pcan.cc
+	${OPTIONAL_TFLITE_DIR}/micro/kernels/filter_bank_log.cc
+)
+
+if("${ARCH}" STREQUAL "xtensa")
+        target_sources(
+        app
+        PRIVATE
+
+        # Xtensa kernels
+        ${OPTIONAL_TFLITE_DIR}/micro/kernels/xtensa/fft_auto_scale_kernel.cc
+        ${OPTIONAL_TFLITE_DIR}/micro/kernels/xtensa/filter_bank_square_root.cc
+        ${OPTIONAL_TFLITE_DIR}/micro/kernels/xtensa/xtensa_square_root.S
+        )
+endif()

samples/modules/tflite-micro/micro_speech/boards/imx8mp_evk_mimx8ml8_adsp.conf

@@ -0,0 +1,8 @@
+CONFIG_LOG_PRINTK=n
+CONFIG_IPM_IMX_MAX_DATA_SIZE_16=n
+CONFIG_IPM_IMX_MAX_DATA_SIZE_4=y
+CONFIG_OPENAMP_WITH_DCACHE=y
+
+#vendor-specific resource table
+CONFIG_OPENAMP_VENDOR_RSC_TABLE=y
+CONFIG_OPENAMP_VENDOR_RSC_TABLE_FILE="nxp_resource_table.c"

samples/modules/tflite-micro/micro_speech/boards/imx8mp_evk_mimx8ml8_adsp.overlay

@@ -0,0 +1,24 @@
+/*
+ * Copyright 2025 NXP
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/ {
+	chosen {
+		/*
+		 * shared memory reserved for the inter-processor communication
+		 */
+		zephyr,ipc_shm = &dspsram3;
+		zephyr,ipc = &mailbox0;
+	};
+
+	dspsram3: memory@942f0000 {
+		compatible = "mmio-sram";
+		reg = <0x942f0000 0x110000>;
+	};
+};
+
+&mailbox0 {
+	status = "okay";
+};

samples/modules/tflite-micro/micro_speech/prj.conf

@@ -0,0 +1,25 @@
+# tflite
+CONFIG_CPP=y
+CONFIG_STD_CPP17=y
+CONFIG_TENSORFLOW_LITE_MICRO=y
+# stack size needed to safely initialize
+# and run TensorFlow Lite Micro operations
+# in the main thread
+CONFIG_MAIN_STACK_SIZE=6144
+CONFIG_REQUIRES_FLOAT_PRINTF=y
+
+# rpmsg
+CONFIG_KERNEL_BIN_NAME="micro_speech_openamp"
+CONFIG_PRINTK=n
+CONFIG_IPM=y
+# rpmsg buffers + OpenAMP overhead + safety margin
+CONFIG_HEAP_MEM_POOL_SIZE=5120
+CONFIG_OPENAMP=y
+CONFIG_OPENAMP_RSC_TABLE_NUM_RPMSG_BUFF=8
+CONFIG_OPENAMP_RSC_TABLE=y
+CONFIG_OPENAMP_MASTER=n
+CONFIG_SHELL=n
+
+# logging
+CONFIG_LOG=y
+CONFIG_LOG_DEFAULT_LEVEL=3

samples/modules/tflite-micro/micro_speech/sample.yaml

@@ -0,0 +1,20 @@
+sample:
+  description: Micro Speech OpenAMP Sample
+  name: micro speech openamp
+common:
+  tags: tensorflow
+  modules:
+    - tflite-micro
+  harness: console
+  harness_config:
+    type: multi_line
+    ordered: false
+    regex:
+      - "Detected: (yes|no|silence|unknown)"
+tests:
+  sample.tensorflow.micro_speech_openamp:
+    platform_allow:
+      - imx8mp_evk/mimx8ml8/adsp
+    integration_platforms:
+      - imx8mp_evk/mimx8ml8/adsp
+    tags: tensorflow, openamp, ipc, audio

samples/modules/tflite-micro/micro_speech/src/inference/audio_preprocessor_int8_model.cpp

@@ -0,0 +1,25 @@
+/*
+ * Copyright 2025 The TensorFlow Authors. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * 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.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#ifndef MICRO_SPEECH_OPENAMP_AUDIO_PREPROCESSOR_INT8_MODEL_H_
+#define MICRO_SPEECH_OPENAMP_AUDIO_PREPROCESSOR_INT8_MODEL_H_
+
+extern const unsigned char g_audio_preprocessor_int8_model[];
+extern const unsigned int g_audio_preprocessor_int8_model_len;
+
+#endif /* MICRO_SPEECH_OPENAMP_AUDIO_PREPROCESSOR_INT8_MODEL_H_ */

samples/modules/tflite-micro/micro_speech/src/inference/audio_preprocessor_int8_model.hpp

@@ -0,0 +1,36 @@
+/*
+ * Copyright 2025 The TensorFlow Authors. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * 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.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#ifndef MICRO_SPEECH_OPENAMP_MODEL_SETTINGS_H_
+#define MICRO_SPEECH_OPENAMP_MODEL_SETTINGS_H_
+
+/* The following values are derived from values used during model training.
+ * If you change the way you preprocess the input, update all these constants.
+ */
+constexpr int kAudioSampleFrequency = 16000;
+constexpr int kFeatureSize = 40;
+constexpr int kFeatureCount = 49;
+constexpr int kFeatureElementCount = (kFeatureSize * kFeatureCount);
+constexpr int kFeatureStrideMs = 20;
+constexpr int kFeatureDurationMs = 30;
+
+/* Variables for the model's output categories. */
+constexpr int kCategoryCount = 4;
+constexpr const char *kCategoryLabels[kCategoryCount] = {"silence", "unknown", "yes", "no"};
+
+#endif /* MICRO_SPEECH_OPENAMP_MODEL_SETTINGS_H_ */