diff --git a/soundtrigger/2.3/default/SoundTriggerHw.cpp b/soundtrigger/2.3/default/SoundTriggerHw.cpp
new file mode 100644
index 0000000..4a39ab5
--- /dev/null
+++ b/soundtrigger/2.3/default/SoundTriggerHw.cpp
@@ -0,0 +1,804 @@
+/*
+ * Copyright (C) 2019 The Android Open Source Project
+ *
+ * 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.
+ */
+
+#define LOG_TAG "SoundTriggerHw"
+
+#include "SoundTriggerHw.h"
+
+#include <android/hidl/allocator/1.0/IAllocator.h>
+#include <android/log.h>
+#include <hidlmemory/mapping.h>
+#include <utility>
+
+using android::hardware::hidl_memory;
+using android::hidl::allocator::V1_0::IAllocator;
+using android::hidl::memory::V1_0::IMemory;
+
+namespace android {
+namespace hardware {
+namespace soundtrigger {
+namespace V2_3 {
+namespace implementation {
+
+/**
+ * According to the HIDL C++ Users Guide: client and server implementations
+ * should never directly refer to anything other than the interface header
+ * generated from the HIDL definition file (ie. ISoundTriggerHw.hal), so
+ * this V2_3 implementation copies the previous implementations and
+ * then adds the new implementation.
+ */
+
+// Begin V2_0 implementation, copied from
+// hardware/interfaces/soundtrigger/2.0/default/SoundTriggerHalImpl.cpp
+
+// static
+void soundModelCallback_(struct sound_trigger_model_event* halEvent, void* cookie) {
+    if (halEvent == NULL) {
+        ALOGW("soundModelCallback called with NULL event");
+        return;
+    }
+    sp<SoundTriggerHw::SoundModelClient> client =
+            wp<SoundTriggerHw::SoundModelClient>(
+                    static_cast<SoundTriggerHw::SoundModelClient*>(cookie))
+                    .promote();
+    if (client == 0) {
+        ALOGW("soundModelCallback called on stale client");
+        return;
+    }
+    if (halEvent->model != client->getHalHandle()) {
+        ALOGW("soundModelCallback call with wrong handle %d on client with handle %d",
+              (int)halEvent->model, (int)client->getHalHandle());
+        return;
+    }
+
+    client->soundModelCallback(halEvent);
+}
+
+// static
+void recognitionCallback_(struct sound_trigger_recognition_event* halEvent, void* cookie) {
+    if (halEvent == NULL) {
+        ALOGW("recognitionCallback call NULL event");
+        return;
+    }
+    sp<SoundTriggerHw::SoundModelClient> client =
+            wp<SoundTriggerHw::SoundModelClient>(
+                    static_cast<SoundTriggerHw::SoundModelClient*>(cookie))
+                    .promote();
+    if (client == 0) {
+        ALOGW("recognitionCallback called on stale client");
+        return;
+    }
+
+    client->recognitionCallback(halEvent);
+}
+
+Return<void> SoundTriggerHw::getProperties(ISoundTriggerHw::getProperties_cb _hidl_cb) {
+    ALOGV("getProperties() mHwDevice %p", mHwDevice);
+    int ret;
+    struct sound_trigger_properties halProperties;
+    ISoundTriggerHw::Properties properties;
+
+    if (mHwDevice == NULL) {
+        ret = -ENODEV;
+        goto exit;
+    }
+
+    ret = mHwDevice->get_properties(mHwDevice, &halProperties);
+
+    convertPropertiesFromHal(&properties, &halProperties);
+
+    ALOGV("getProperties implementor %s recognitionModes %08x", properties.implementor.c_str(),
+          properties.recognitionModes);
+
+exit:
+    _hidl_cb(ret, properties);
+    return Void();
+}
+
+int SoundTriggerHw::doLoadSoundModel(const V2_0::ISoundTriggerHw::SoundModel& soundModel,
+                                     sp<SoundTriggerHw::SoundModelClient> client) {
+    int32_t ret = 0;
+    struct sound_trigger_sound_model* halSoundModel;
+
+    ALOGV("doLoadSoundModel() data size %zu", soundModel.data.size());
+
+    if (mHwDevice == NULL) {
+        ret = -ENODEV;
+        goto exit;
+    }
+
+    halSoundModel = convertSoundModelToHal(&soundModel);
+    if (halSoundModel == NULL) {
+        ret = -EINVAL;
+        goto exit;
+    }
+
+    sound_model_handle_t halHandle;
+    ret = mHwDevice->load_sound_model(mHwDevice, halSoundModel, soundModelCallback_, client.get(),
+                                      &halHandle);
+
+    free(halSoundModel);
+
+    if (ret != 0) {
+        goto exit;
+    }
+
+    client->setHalHandle(halHandle);
+    {
+        AutoMutex lock(mLock);
+        mClients.add(client->getId(), client);
+    }
+
+exit:
+    return ret;
+}
+
+Return<void> SoundTriggerHw::loadSoundModel(const V2_0::ISoundTriggerHw::SoundModel& soundModel,
+                                            const sp<V2_0::ISoundTriggerHwCallback>& callback,
+                                            V2_0::ISoundTriggerHwCallback::CallbackCookie cookie,
+                                            ISoundTriggerHw::loadSoundModel_cb _hidl_cb) {
+    sp<SoundTriggerHw::SoundModelClient> client =
+            new SoundModelClient_2_0(nextUniqueModelId(), cookie, callback);
+    _hidl_cb(doLoadSoundModel(soundModel, client), client->getId());
+    return Void();
+}
+
+Return<void> SoundTriggerHw::loadPhraseSoundModel(
+        const V2_0::ISoundTriggerHw::PhraseSoundModel& soundModel,
+        const sp<V2_0::ISoundTriggerHwCallback>& callback,
+        V2_0::ISoundTriggerHwCallback::CallbackCookie cookie,
+        ISoundTriggerHw::loadPhraseSoundModel_cb _hidl_cb) {
+    sp<SoundTriggerHw::SoundModelClient> client =
+            new SoundModelClient_2_0(nextUniqueModelId(), cookie, callback);
+    _hidl_cb(doLoadSoundModel((const V2_0::ISoundTriggerHw::SoundModel&)soundModel, client),
+             client->getId());
+    return Void();
+}
+
+Return<int32_t> SoundTriggerHw::unloadSoundModel(int32_t modelHandle) {
+    int32_t ret;
+    sp<SoundTriggerHw::SoundModelClient> client;
+
+    if (mHwDevice == NULL) {
+        ret = -ENODEV;
+        goto exit;
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            ret = -ENOSYS;
+            goto exit;
+        }
+    }
+
+    ret = mHwDevice->unload_sound_model(mHwDevice, client->getHalHandle());
+
+    mClients.removeItem(modelHandle);
+
+exit:
+    return ret;
+}
+
+Return<int32_t> SoundTriggerHw::startRecognition(
+        int32_t modelHandle, const V2_0::ISoundTriggerHw::RecognitionConfig& config,
+        const sp<V2_0::ISoundTriggerHwCallback>& /* callback */, int32_t /* cookie */) {
+    int32_t ret;
+    sp<SoundTriggerHw::SoundModelClient> client;
+    struct sound_trigger_recognition_config* halConfig;
+
+    if (mHwDevice == NULL) {
+        ret = -ENODEV;
+        goto exit;
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            ret = -ENOSYS;
+            goto exit;
+        }
+    }
+
+    halConfig =
+            convertRecognitionConfigToHal((const V2_0::ISoundTriggerHw::RecognitionConfig*)&config);
+
+    if (halConfig == NULL) {
+        ret = -EINVAL;
+        goto exit;
+    }
+    ret = mHwDevice->start_recognition(mHwDevice, client->getHalHandle(), halConfig,
+                                       recognitionCallback_, client.get());
+
+    free(halConfig);
+
+exit:
+    return ret;
+}
+
+Return<int32_t> SoundTriggerHw::stopRecognition(int32_t modelHandle) {
+    int32_t ret;
+    sp<SoundTriggerHw::SoundModelClient> client;
+    if (mHwDevice == NULL) {
+        ret = -ENODEV;
+        goto exit;
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            ret = -ENOSYS;
+            goto exit;
+        }
+    }
+
+    ret = mHwDevice->stop_recognition(mHwDevice, client->getHalHandle());
+
+exit:
+    return ret;
+}
+
+Return<int32_t> SoundTriggerHw::stopAllRecognitions() {
+    int32_t ret;
+    if (mHwDevice == NULL) {
+        ret = -ENODEV;
+        goto exit;
+    }
+
+    ret = mHwDevice->stop_all_recognitions(mHwDevice);
+
+exit:
+    return ret;
+}
+
+SoundTriggerHw::SoundTriggerHw() : mModuleName("primary"), mHwDevice(NULL), mNextModelId(1) {}
+
+void SoundTriggerHw::onFirstRef() {
+    const hw_module_t* mod;
+    int rc;
+
+    rc = hw_get_module_by_class(SOUND_TRIGGER_HARDWARE_MODULE_ID, mModuleName, &mod);
+    if (rc != 0) {
+        ALOGE("couldn't load sound trigger module %s.%s (%s)", SOUND_TRIGGER_HARDWARE_MODULE_ID,
+              mModuleName, strerror(-rc));
+        return;
+    }
+    rc = sound_trigger_hw_device_open(mod, &mHwDevice);
+    if (rc != 0) {
+        ALOGE("couldn't open sound trigger hw device in %s.%s (%s)",
+              SOUND_TRIGGER_HARDWARE_MODULE_ID, mModuleName, strerror(-rc));
+        mHwDevice = NULL;
+        return;
+    }
+    if (mHwDevice->common.version != SOUND_TRIGGER_DEVICE_API_VERSION_1_3) {
+        ALOGE("wrong sound trigger hw device version %04x", mHwDevice->common.version);
+        sound_trigger_hw_device_close(mHwDevice);
+        mHwDevice = NULL;
+        return;
+    }
+
+    ALOGI("onFirstRef() mModuleName %s mHwDevice %p", mModuleName, mHwDevice);
+}
+
+SoundTriggerHw::~SoundTriggerHw() {
+    if (mHwDevice != NULL) {
+        sound_trigger_hw_device_close(mHwDevice);
+    }
+}
+
+uint32_t SoundTriggerHw::nextUniqueModelId() {
+    uint32_t modelId = 0;
+    {
+        AutoMutex lock(mLock);
+        do {
+            modelId = atomic_fetch_add_explicit(&mNextModelId, (uint_fast32_t)1,
+                                                memory_order_acq_rel);
+        } while (mClients.valueFor(modelId) != 0 && modelId != 0);
+    }
+    LOG_ALWAYS_FATAL_IF(modelId == 0, "wrap around in sound model IDs, num loaded models %zu",
+                        mClients.size());
+    return modelId;
+}
+
+void SoundTriggerHw::convertUuidFromHal(Uuid* uuid, const sound_trigger_uuid_t* halUuid) {
+    uuid->timeLow = halUuid->timeLow;
+    uuid->timeMid = halUuid->timeMid;
+    uuid->versionAndTimeHigh = halUuid->timeHiAndVersion;
+    uuid->variantAndClockSeqHigh = halUuid->clockSeq;
+    memcpy(&uuid->node[0], &halUuid->node[0], 6);
+}
+
+void SoundTriggerHw::convertUuidToHal(sound_trigger_uuid_t* halUuid, const Uuid* uuid) {
+    halUuid->timeLow = uuid->timeLow;
+    halUuid->timeMid = uuid->timeMid;
+    halUuid->timeHiAndVersion = uuid->versionAndTimeHigh;
+    halUuid->clockSeq = uuid->variantAndClockSeqHigh;
+    memcpy(&halUuid->node[0], &uuid->node[0], 6);
+}
+
+void SoundTriggerHw::convertPropertiesFromHal(
+        ISoundTriggerHw::Properties* properties,
+        const struct sound_trigger_properties* halProperties) {
+    properties->implementor = halProperties->implementor;
+    properties->description = halProperties->description;
+    properties->version = halProperties->version;
+    convertUuidFromHal(&properties->uuid, &halProperties->uuid);
+    properties->maxSoundModels = halProperties->max_sound_models;
+    properties->maxKeyPhrases = halProperties->max_key_phrases;
+    properties->maxUsers = halProperties->max_users;
+    properties->recognitionModes = halProperties->recognition_modes;
+    properties->captureTransition = halProperties->capture_transition;
+    properties->maxBufferMs = halProperties->max_buffer_ms;
+    properties->concurrentCapture = halProperties->concurrent_capture;
+    properties->triggerInEvent = halProperties->trigger_in_event;
+    properties->powerConsumptionMw = halProperties->power_consumption_mw;
+}
+
+void SoundTriggerHw::convertTriggerPhraseToHal(struct sound_trigger_phrase* halTriggerPhrase,
+                                               const ISoundTriggerHw::Phrase* triggerPhrase) {
+    halTriggerPhrase->id = triggerPhrase->id;
+    halTriggerPhrase->recognition_mode = triggerPhrase->recognitionModes;
+    unsigned int i;
+
+    halTriggerPhrase->num_users =
+            std::min((int)triggerPhrase->users.size(), SOUND_TRIGGER_MAX_USERS);
+    for (i = 0; i < halTriggerPhrase->num_users; i++) {
+        halTriggerPhrase->users[i] = triggerPhrase->users[i];
+    }
+
+    strlcpy(halTriggerPhrase->locale, triggerPhrase->locale.c_str(), SOUND_TRIGGER_MAX_LOCALE_LEN);
+    strlcpy(halTriggerPhrase->text, triggerPhrase->text.c_str(), SOUND_TRIGGER_MAX_STRING_LEN);
+}
+
+struct sound_trigger_sound_model* SoundTriggerHw::convertSoundModelToHal(
+        const V2_0::ISoundTriggerHw::SoundModel* soundModel) {
+    struct sound_trigger_sound_model* halModel = NULL;
+    if (soundModel->type == V2_0::SoundModelType::KEYPHRASE) {
+        size_t allocSize =
+                sizeof(struct sound_trigger_phrase_sound_model) + soundModel->data.size();
+        struct sound_trigger_phrase_sound_model* halKeyPhraseModel =
+                static_cast<struct sound_trigger_phrase_sound_model*>(malloc(allocSize));
+        LOG_ALWAYS_FATAL_IF(halKeyPhraseModel == NULL,
+                            "malloc failed for size %zu in convertSoundModelToHal PHRASE",
+                            allocSize);
+
+        const V2_0::ISoundTriggerHw::PhraseSoundModel* keyPhraseModel =
+                reinterpret_cast<const V2_0::ISoundTriggerHw::PhraseSoundModel*>(soundModel);
+
+        size_t i;
+        for (i = 0; i < keyPhraseModel->phrases.size() && i < SOUND_TRIGGER_MAX_PHRASES; i++) {
+            convertTriggerPhraseToHal(&halKeyPhraseModel->phrases[i], &keyPhraseModel->phrases[i]);
+        }
+        halKeyPhraseModel->num_phrases = (unsigned int)i;
+        halModel = reinterpret_cast<struct sound_trigger_sound_model*>(halKeyPhraseModel);
+        halModel->data_offset = sizeof(struct sound_trigger_phrase_sound_model);
+    } else {
+        size_t allocSize = sizeof(struct sound_trigger_sound_model) + soundModel->data.size();
+        halModel = static_cast<struct sound_trigger_sound_model*>(malloc(allocSize));
+        LOG_ALWAYS_FATAL_IF(halModel == NULL,
+                            "malloc failed for size %zu in convertSoundModelToHal GENERIC",
+                            allocSize);
+
+        halModel->data_offset = sizeof(struct sound_trigger_sound_model);
+    }
+    halModel->type = (sound_trigger_sound_model_type_t)soundModel->type;
+    convertUuidToHal(&halModel->uuid, &soundModel->uuid);
+    convertUuidToHal(&halModel->vendor_uuid, &soundModel->vendorUuid);
+    halModel->data_size = soundModel->data.size();
+    uint8_t* dst = reinterpret_cast<uint8_t*>(halModel) + halModel->data_offset;
+    const uint8_t* src = reinterpret_cast<const uint8_t*>(&soundModel->data[0]);
+    memcpy(dst, src, soundModel->data.size());
+
+    return halModel;
+}
+
+void SoundTriggerHw::convertPhraseRecognitionExtraToHal(
+        struct sound_trigger_phrase_recognition_extra* halExtra,
+        const V2_0::PhraseRecognitionExtra* extra) {
+    halExtra->id = extra->id;
+    halExtra->recognition_modes = extra->recognitionModes;
+    halExtra->confidence_level = extra->confidenceLevel;
+
+    unsigned int i;
+    for (i = 0; i < extra->levels.size() && i < SOUND_TRIGGER_MAX_USERS; i++) {
+        halExtra->levels[i].user_id = extra->levels[i].userId;
+        halExtra->levels[i].level = extra->levels[i].levelPercent;
+    }
+    halExtra->num_levels = i;
+}
+
+struct sound_trigger_recognition_config* SoundTriggerHw::convertRecognitionConfigToHal(
+        const V2_0::ISoundTriggerHw::RecognitionConfig* config) {
+    size_t allocSize = sizeof(struct sound_trigger_recognition_config) + config->data.size();
+    struct sound_trigger_recognition_config* halConfig =
+            static_cast<struct sound_trigger_recognition_config*>(malloc(allocSize));
+
+    LOG_ALWAYS_FATAL_IF(halConfig == NULL,
+                        "malloc failed for size %zu in convertRecognitionConfigToHal", allocSize);
+
+    halConfig->capture_handle = (audio_io_handle_t)config->captureHandle;
+    halConfig->capture_device = (audio_devices_t)config->captureDevice;
+    halConfig->capture_requested = config->captureRequested;
+
+    unsigned int i;
+    for (i = 0; i < config->phrases.size() && i < SOUND_TRIGGER_MAX_PHRASES; i++) {
+        convertPhraseRecognitionExtraToHal(&halConfig->phrases[i], &config->phrases[i]);
+    }
+    halConfig->num_phrases = i;
+
+    halConfig->data_offset = sizeof(struct sound_trigger_recognition_config);
+    halConfig->data_size = config->data.size();
+    uint8_t* dst = reinterpret_cast<uint8_t*>(halConfig) + halConfig->data_offset;
+    const uint8_t* src = reinterpret_cast<const uint8_t*>(&config->data[0]);
+    memcpy(dst, src, config->data.size());
+    return halConfig;
+}
+
+// static
+void SoundTriggerHw::convertSoundModelEventFromHal(
+        V2_0::ISoundTriggerHwCallback::ModelEvent* event,
+        const struct sound_trigger_model_event* halEvent) {
+    event->status = (V2_0::ISoundTriggerHwCallback::SoundModelStatus)halEvent->status;
+    // event->model to be remapped by called
+    event->data.setToExternal(const_cast<uint8_t*>(reinterpret_cast<const uint8_t*>(halEvent)) +
+                                      halEvent->data_offset,
+                              halEvent->data_size);
+}
+
+// static
+void SoundTriggerHw::convertPhaseRecognitionEventFromHal(
+        V2_0::ISoundTriggerHwCallback::PhraseRecognitionEvent* event,
+        const struct sound_trigger_phrase_recognition_event* halEvent) {
+    event->phraseExtras.resize(halEvent->num_phrases);
+    for (unsigned int i = 0; i < halEvent->num_phrases; i++) {
+        convertPhraseRecognitionExtraFromHal(&event->phraseExtras[i], &halEvent->phrase_extras[i]);
+    }
+    convertRecognitionEventFromHal(&event->common, &halEvent->common);
+}
+
+// static
+void SoundTriggerHw::convertRecognitionEventFromHal(
+        V2_0::ISoundTriggerHwCallback::RecognitionEvent* event,
+        const struct sound_trigger_recognition_event* halEvent) {
+    event->status = static_cast<V2_0::ISoundTriggerHwCallback::RecognitionStatus>(halEvent->status);
+    event->type = static_cast<V2_0::SoundModelType>(halEvent->type);
+    // event->model to be remapped by called
+    event->captureAvailable = halEvent->capture_available;
+    event->captureSession = halEvent->capture_session;
+    event->captureDelayMs = halEvent->capture_delay_ms;
+    event->capturePreambleMs = halEvent->capture_preamble_ms;
+    event->triggerInData = halEvent->trigger_in_data;
+    event->audioConfig.sampleRateHz = halEvent->audio_config.sample_rate;
+    event->audioConfig.channelMask =
+            (audio::common::V2_0::AudioChannelMask)halEvent->audio_config.channel_mask;
+    event->audioConfig.format = (audio::common::V2_0::AudioFormat)halEvent->audio_config.format;
+    event->data.setToExternal(const_cast<uint8_t*>(reinterpret_cast<const uint8_t*>(halEvent)) +
+                                      halEvent->data_offset,
+                              halEvent->data_size);
+}
+
+// static
+void SoundTriggerHw::convertPhraseRecognitionExtraFromHal(
+        V2_0::PhraseRecognitionExtra* extra,
+        const struct sound_trigger_phrase_recognition_extra* halExtra) {
+    extra->id = halExtra->id;
+    extra->recognitionModes = halExtra->recognition_modes;
+    extra->confidenceLevel = halExtra->confidence_level;
+
+    extra->levels.resize(halExtra->num_levels);
+    for (unsigned int i = 0; i < halExtra->num_levels; i++) {
+        extra->levels[i].userId = halExtra->levels[i].user_id;
+        extra->levels[i].levelPercent = halExtra->levels[i].level;
+    }
+}
+
+void SoundTriggerHw::SoundModelClient_2_0::recognitionCallback(
+        struct sound_trigger_recognition_event* halEvent) {
+    if (halEvent->type == SOUND_MODEL_TYPE_KEYPHRASE) {
+        V2_0::ISoundTriggerHwCallback::PhraseRecognitionEvent event;
+        convertPhaseRecognitionEventFromHal(
+                &event, reinterpret_cast<sound_trigger_phrase_recognition_event*>(halEvent));
+        event.common.model = mId;
+        mCallback->phraseRecognitionCallback(event, mCookie);
+    } else {
+        V2_0::ISoundTriggerHwCallback::RecognitionEvent event;
+        convertRecognitionEventFromHal(&event, halEvent);
+        event.model = mId;
+        mCallback->recognitionCallback(event, mCookie);
+    }
+}
+
+void SoundTriggerHw::SoundModelClient_2_0::soundModelCallback(
+        struct sound_trigger_model_event* halEvent) {
+    V2_0::ISoundTriggerHwCallback::ModelEvent event;
+    convertSoundModelEventFromHal(&event, halEvent);
+    event.model = mId;
+    mCallback->soundModelCallback(event, mCookie);
+}
+
+// Begin V2_1 implementation, copied from
+// hardware/interfaces/soundtrigger/2.1/default/SoundTriggerHw.cpp
+
+namespace {
+
+// Backs up by the vector with the contents of shared memory.
+// It is assumed that the passed hidl_vector is empty, so it's
+// not cleared if the memory is a null object.
+// The caller needs to keep the returned sp<IMemory> as long as
+// the data is needed.
+std::pair<bool, sp<IMemory>> memoryAsVector(const hidl_memory& m, hidl_vec<uint8_t>* vec) {
+    sp<IMemory> memory;
+    if (m.size() == 0) {
+        return std::make_pair(true, memory);
+    }
+    memory = mapMemory(m);
+    if (memory != nullptr) {
+        memory->read();
+        vec->setToExternal(static_cast<uint8_t*>(static_cast<void*>(memory->getPointer())),
+                           memory->getSize());
+        return std::make_pair(true, memory);
+    }
+    ALOGE("%s: Could not map HIDL memory to IMemory", __func__);
+    return std::make_pair(false, memory);
+}
+
+// Moves the data from the vector into allocated shared memory,
+// emptying the vector.
+// It is assumed that the passed hidl_memory is a null object, so it's
+// not reset if the vector is empty.
+// The caller needs to keep the returned sp<IMemory> as long as
+// the data is needed.
+std::pair<bool, sp<IMemory>> moveVectorToMemory(hidl_vec<uint8_t>* v, hidl_memory* mem) {
+    sp<IMemory> memory;
+    if (v->size() == 0) {
+        return std::make_pair(true, memory);
+    }
+    sp<IAllocator> ashmem = IAllocator::getService("ashmem");
+    if (ashmem == 0) {
+        ALOGE("Failed to retrieve ashmem allocator service");
+        return std::make_pair(false, memory);
+    }
+    bool success = false;
+    Return<void> r = ashmem->allocate(v->size(), [&](bool s, const hidl_memory& m) {
+        success = s;
+        if (success) *mem = m;
+    });
+    if (r.isOk() && success) {
+        memory = hardware::mapMemory(*mem);
+        if (memory != 0) {
+            memory->update();
+            memcpy(memory->getPointer(), v->data(), v->size());
+            memory->commit();
+            v->resize(0);
+            return std::make_pair(true, memory);
+        } else {
+            ALOGE("Failed to map allocated ashmem");
+        }
+    } else {
+        ALOGE("Failed to allocate %llu bytes from ashmem", (unsigned long long)v->size());
+    }
+    return std::make_pair(false, memory);
+}
+
+}  // namespace
+
+Return<void> SoundTriggerHw::loadSoundModel_2_1(
+        const V2_1::ISoundTriggerHw::SoundModel& soundModel,
+        const sp<V2_1::ISoundTriggerHwCallback>& callback, int32_t cookie,
+        V2_1::ISoundTriggerHw::loadSoundModel_2_1_cb _hidl_cb) {
+    // It is assumed that legacy data vector is empty, thus making copy is cheap.
+    V2_0::ISoundTriggerHw::SoundModel soundModel_2_0(soundModel.header);
+    auto result = memoryAsVector(soundModel.data, &soundModel_2_0.data);
+    if (result.first) {
+        sp<SoundModelClient> client =
+                new SoundModelClient_2_1(nextUniqueModelId(), cookie, callback);
+        _hidl_cb(doLoadSoundModel(soundModel_2_0, client), client->getId());
+        return Void();
+    }
+    _hidl_cb(-ENOMEM, 0);
+    return Void();
+}
+
+Return<void> SoundTriggerHw::loadPhraseSoundModel_2_1(
+        const V2_1::ISoundTriggerHw::PhraseSoundModel& soundModel,
+        const sp<V2_1::ISoundTriggerHwCallback>& callback, int32_t cookie,
+        V2_1::ISoundTriggerHw::loadPhraseSoundModel_2_1_cb _hidl_cb) {
+    V2_0::ISoundTriggerHw::PhraseSoundModel soundModel_2_0;
+    // It is assumed that legacy data vector is empty, thus making copy is cheap.
+    soundModel_2_0.common = soundModel.common.header;
+    // Avoid copying phrases data.
+    soundModel_2_0.phrases.setToExternal(
+            const_cast<V2_0::ISoundTriggerHw::Phrase*>(soundModel.phrases.data()),
+            soundModel.phrases.size());
+    auto result = memoryAsVector(soundModel.common.data, &soundModel_2_0.common.data);
+    if (result.first) {
+        sp<SoundModelClient> client =
+                new SoundModelClient_2_1(nextUniqueModelId(), cookie, callback);
+        _hidl_cb(doLoadSoundModel((const V2_0::ISoundTriggerHw::SoundModel&)soundModel_2_0, client),
+                 client->getId());
+        return Void();
+    }
+    _hidl_cb(-ENOMEM, 0);
+    return Void();
+}
+
+Return<int32_t> SoundTriggerHw::startRecognition_2_1(
+        int32_t modelHandle, const V2_1::ISoundTriggerHw::RecognitionConfig& config,
+        const sp<V2_1::ISoundTriggerHwCallback>& callback, int32_t cookie) {
+    // It is assumed that legacy data vector is empty, thus making copy is cheap.
+    V2_0::ISoundTriggerHw::RecognitionConfig config_2_0(config.header);
+    auto result = memoryAsVector(config.data, &config_2_0.data);
+    return result.first ? startRecognition(modelHandle, config_2_0, callback, cookie)
+                        : Return<int32_t>(-ENOMEM);
+}
+
+void SoundTriggerHw::SoundModelClient_2_1::recognitionCallback(
+        struct sound_trigger_recognition_event* halEvent) {
+    if (halEvent->type == SOUND_MODEL_TYPE_KEYPHRASE) {
+        V2_0::ISoundTriggerHwCallback::PhraseRecognitionEvent event_2_0;
+        convertPhaseRecognitionEventFromHal(
+                &event_2_0, reinterpret_cast<sound_trigger_phrase_recognition_event*>(halEvent));
+        event_2_0.common.model = mId;
+        V2_1::ISoundTriggerHwCallback::PhraseRecognitionEvent event;
+        event.phraseExtras.setToExternal(event_2_0.phraseExtras.data(),
+                                         event_2_0.phraseExtras.size());
+        auto result = moveVectorToMemory(&event_2_0.common.data, &event.common.data);
+        if (result.first) {
+            // The data vector is now empty, thus copying is cheap.
+            event.common.header = event_2_0.common;
+            mCallback->phraseRecognitionCallback_2_1(event, mCookie);
+        }
+    } else {
+        V2_1::ISoundTriggerHwCallback::RecognitionEvent event;
+        convertRecognitionEventFromHal(&event.header, halEvent);
+        event.header.model = mId;
+        auto result = moveVectorToMemory(&event.header.data, &event.data);
+        if (result.first) {
+            mCallback->recognitionCallback_2_1(event, mCookie);
+        }
+    }
+}
+
+void SoundTriggerHw::SoundModelClient_2_1::soundModelCallback(
+        struct sound_trigger_model_event* halEvent) {
+    V2_1::ISoundTriggerHwCallback::ModelEvent event;
+    convertSoundModelEventFromHal(&event.header, halEvent);
+    event.header.model = mId;
+    auto result = moveVectorToMemory(&event.header.data, &event.data);
+    if (result.first) {
+        mCallback->soundModelCallback_2_1(event, mCookie);
+    }
+}
+
+// Begin V2_2 implementation, copied from
+// hardware/interfaces/soundtrigger/2.2/default/SoundTriggerHw.cpp
+
+Return<int32_t> SoundTriggerHw::getModelState(int32_t modelHandle) {
+    sp<SoundModelClient> client;
+    if (mHwDevice == NULL) {
+        return -ENODEV;
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            return -ENOSYS;
+        }
+    }
+
+    return mHwDevice->get_model_state(mHwDevice, client->getHalHandle());
+}
+
+// Begin V2_3 implementation
+
+Return<int32_t> SoundTriggerHw::setParameter(V2_0::SoundModelHandle modelHandle,
+                                             ModelParameter modelParam, int32_t value) {
+    sp<SoundModelClient> client;
+    if (mHwDevice == NULL) {
+        return -ENODEV;
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            return -EINVAL;
+        }
+    }
+
+    return mHwDevice->set_parameter(mHwDevice, client->getHalHandle(),
+                                    convertModelParameterToHal(modelParam), value);
+}
+
+Return<void> SoundTriggerHw::getParameter(V2_0::SoundModelHandle modelHandle,
+                                          ModelParameter modelParam, getParameter_cb _hidl_cb) {
+    sp<SoundModelClient> client;
+    if (mHwDevice == NULL) {
+        _hidl_cb(-ENODEV, 0);
+        return Void();
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            _hidl_cb(-EINVAL, 0);
+            return Void();
+        }
+    }
+
+    int32_t value;
+    int32_t status = mHwDevice->get_parameter(mHwDevice, client->getHalHandle(),
+                                              convertModelParameterToHal(modelParam), &value);
+    _hidl_cb(status, value);
+    return Void();
+}
+
+Return<void> SoundTriggerHw::queryParameter(V2_0::SoundModelHandle modelHandle,
+                                            ModelParameter modelParam, queryParameter_cb _hidl_cb) {
+    OptionalModelParameterRange optionalParamRange;
+    sp<SoundModelClient> client;
+    if (mHwDevice == NULL) {
+        _hidl_cb(-ENODEV, optionalParamRange);
+        return Void();
+    }
+
+    {
+        AutoMutex lock(mLock);
+        client = mClients.valueFor(modelHandle);
+        if (client == 0) {
+            _hidl_cb(-EINVAL, optionalParamRange);
+            return Void();
+        }
+    }
+
+    sound_trigger_model_parameter_range_t paramRange;
+    int32_t status = mHwDevice->query_parameter(
+            mHwDevice, client->getHalHandle(), convertModelParameterToHal(modelParam), &paramRange);
+
+    if (status == 0) {
+        optionalParamRange.range({.start = paramRange.start, .end = paramRange.end});
+    }
+    _hidl_cb(status, optionalParamRange);
+    return Void();
+}
+
+// static
+sound_trigger_model_parameter_t SoundTriggerHw::convertModelParameterToHal(ModelParameter param) {
+    switch (param) {
+        case ModelParameter::THRESHOLD_FACTOR:
+            return MODEL_PARAMETER_THRESHOLD_FACTOR;
+        case ModelParameter::INVALID:
+        default:
+            return MODEL_PARAMETER_INVALID;
+    }
+}
+
+// Methods from ::android::hidl::base::V1_0::IBase follow.
+
+ISoundTriggerHw* HIDL_FETCH_ISoundTriggerHw(const char* /* name */) {
+    return new SoundTriggerHw();
+}
+
+}  // namespace implementation
+}  // namespace V2_3
+}  // namespace soundtrigger
+}  // namespace hardware
+}  // namespace android