summaryrefslogtreecommitdiffstats
path: root/recipes-demo-hmi/libqtappfw/libqtappfw_git.bb
blob: ebc9ce9d4b021f17d96c639c368259f5de95c4a3 (plain) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

SUMMARY     = "AGL Qt AppFW Library"
DESCRIPTION = "libqtappfw"
HOMEPAGE    = "http://docs.automotivelinux.org"
LICENSE     = "Apache-2.0"
SECTION     = "libs"

BBCLASSEXTEND = "nativesdk"

LIC_FILES_CHKSUM = "file://LICENSE;md5=ae6497158920d9524cf208c09cc4c984"

DEPENDS += "qtbase qtdeclarative qtwebsockets"

inherit cmake_qt5

SRC_URI = "git://gerrit.automotivelinux.org/gerrit/src/libqtappfw;protocol=https;branch=${AGL_BRANCH}"
SRCREV  = "da091649e775b23a35eb88df1c2cf9ce3fb350f7"
S       = "${WORKDIR}/git/"

# PV needs to be modified with SRCPV to work AUTOREV correctly
PV = "0.0+git${SRCPV}"
class="na">#test-vosk) - [Test Snips](#test-snips) - [Supported Targets](#supported-targets) - [Maintainers](#maintainers) ## Introduction The `meta-offline-voice-agent` layer integrates the Vosk API and Snips (Inference Only) to provide offline speech recognition and command execution for Automotive Grade Linux. The layer is based on the Kaldi ASR Toolkit, which allows for accurate and efficient speech recognition in the AGL platform, and Snips which provides us with a lightweight Natural Language Intent Engine. ## Layer Status **Status**: *WIP (Work In Progress)* This layer is currently in development and integrates the Vosk, Snips (Inference Only), and RASA libraries. Speech recognition has been verified using test scripts from the [vosk-api python examples](https://github.com/alphacep/vosk-api/tree/master/python/example). Additionally, Snips has undergone testing and verification to ensure its functionality. Currently, work on integrating RASA and command execution is still in progress. ## External Dependencies This layer does not have any external layer dependency. ## Working Features The following features are currently working in the `meta-offline-voice-agent` layer: - [Vosk API (Python)](https://github.com/alphacep/vosk-api/tree/master/python) - [Vosk Websocket Server](https://github.com/alphacep/vosk-server/tree/master/websocket) - [Snips Inference](https://github.com/malik727/snips-inference-agl) ## Testing Features on AGL ### Build Layer In order to test the features of this layer you first need to build it as part of your final AGL image. First of all ensure that you have all the external layer dependencies included. Then you can use the following set of commands to initialize and build this layer into the `agl-demo-platform` qemux86_64 image: ```shell $ source master/meta-agl/scripts/aglsetup.sh -m qemux86-64 -b build-master agl-demo agl-devel agl-offline-voice-agent $ source agl-init-build-env $ bitbake agl-demo-platform ``` The build can take anywhere from 6 hours to 24 hours or even more depending upon compute power of your machine. After the build completes you can use the following command to boot into your AGL image: (you need to install QEMU if not already for the command to work) ```shell $ runqemu tmp/deploy/images/qemux86-64/agl-demo-platform-qemux86-64.qemuboot.conf kvm serialstdio slirp publicvnc audio ``` ### Test Vosk (**Not Recommended**) The simplest way to test Vosk API is by using the following command: ```shell $ ptest-runner python3-vosk-api ``` In order for the above command to work you need to turn on `ptests` by adding the following lines to your `local.conf` that can be found at `meta-agl-devel/templates/feature/agl-offline-voice-agent/50_local.conf.inc`: ```shell DISTRO_FEATURES:append = " ptest" EXTRA_IMAGE_FEATURES += "ptest-pkgs" ``` The above method may be the easiest one but it's not recommended because `ptests` increase the image build times by a substantial amount. You can look into the official [vosk-api docs](https://alphacephei.com/vosk/install) for usage and other ways of testing. ### Test Snips (**Important**) Currently, there are some library linking issues between NumPy, SciPy, and OpenBLAS. While we investigate and fix them you need to use `LD_PRELOAD` method as a workaround for Snips to work properly. Input the following command as soon as you boot into the target image: ```shell $ export LD_PRELOAD=/usr/lib/libopenblas.so.0 ``` In order to test the Snips NLU Intent Engine you can use the sample [pre-trained model](https://github.com/malik727/snips-model-agl), by default it automatically gets built into the target image when you include this layer. To perform inference using this model you can run the following command inside your target image: ```shell $ snips-inference parse /usr/share/nlu/snips/model/ -q "your command here" ``` This is just a sample model and may not be able to handle all types of commands. You can always train your own intent engine model using your custom dataset, for more details on how to do that you can look into the README files of [snips-sdk-agl](https://github.com/malik727/snips-sdk-agl), [snips-model-agl](https://github.com/malik727/snips-model-agl), and [snips-inference-agl](https://github.com/malik727/snips-inference-agl). ## Supported Targets Currently, the following targets are fully supported: - QEMU x86-64 (work in progress) ## Maintainers - Malik Talha <talhamalik727x@gmail.com> - Aman Arora <aman.arora9848@gmail.com>