DigitalPhonetics / adviser

ADvISER is a flexible framework to encourage task-oriented dialog system research & development

 

Documentation

Please see the documentation for more details.

Installation

Note: Adviser 2.0 is currently only tested on Linux and Mac (for M1 chips see the extra section near the bottom of this file). (Windows is possible using WSL2 or check the instructions at the bottom for an experimental Windows setup)

Downloading the code

If Git is not installated on your machine, just download the Adviser 2.0 file available in relases section. Then unzip and navigate to the main folder. Note that this method has some disadvantages (you'll only be able to run basic text-to-text terminal conversations).

Cloning the repository (recommended)

If Git is installed on your machine, you may instead clone the repository by entering in a terminal window:

git clone https://github.com/DigitalPhonetics/adviser.git

System Library Requirements

• If you want to use speech in-/output, please make sure you have the hdf5, portaudio and sndfile libraries installed.
• If you want to make use of the function services.service.Service.draw_system_graph,

you will need to install the graphviz library via your system's package manager. If you can't install it (no sufficient user rights), don't use this function in your scripts.

On Ubuntu e.g.:

sudo apt-get install graphviz

On Mac, you will need to install homebrew by executing:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install.sh)"

and then calling brew install graphviz.

For other OS please see https://graphviz.gitlab.io/download/.

Install python requirements with pip

ADvISER needs to be executed in a Python3 environment.

Once you obtained the code, navigate to its top level directory where you will find the file requirements_base.txt, which lists all modules you need to run a basic text-to-text version of ADvISER. We suggest to create a virtual environment from the top level directory, as shown below, followed by installing the necessary packages.

1. (Requires pip or pip3) Make sure you have virtualenv installed by executing

python3 -m pip install --user virtualenv

2. Create the virtual environment (replace envname with a name of your choice)

python3 -m venv <path-to-env>

3. Source the environment (this has to be repeated every time you want to use ADVISER inside a new terminal session)

source <path-to-env>/bin/activate

4. Install the required packages

pip install -r requirements_base.txt

(NOTE: or requirements_multimodal.txt if you want to use ASR / TTS)

5. Navigate to the adviser folder

cd adviser

and, to make sure your installation is working, execute

python run_chat.py lecturers

You can type text to chat with the system (confirm your utterance by pressing the Enter-Key once) or type bye (followed by pressing the Enter-Key once) to end the conversation.

To see more of the available options, run

python run_chat.py --help

6. OPTIONAL: If you want to use multimodal functionallity, e.g. ASR / TTS/ ..., download the models via the script download_models.sh found in the top level folder

sh download_models.sh

NOTE: this also requires you to install requirements_multimodal.txt in step 4.

You can enable ASR / TTS by adding --asr and --tts to the command line options of run_chat.py (NOTE: for TTS, we recommend you run the code on a CUDA-enabled device and append --cuda to the command line options for drastic performance increase).

7. OPTIONAL: If you want to run the demo with all services enabled, please make sure you executed step 6 and installed the requirements_multimodal.txt. Then, additional requirements must be compiled by yourself - follow the guide in tools/OpenFace/how_to_install.md for this.

Then, try running

python run_demo_multidomain.py

Instructions for Macs with M1 Chips

In general, everything should work if you're using conda instead of pip. For pip users, the following installation instructions worked:

1. Install the system library requirements as stated above (using homebrew).
2. Install additional reqiuirements: brew install rust and brew install portaudio

2. pip install -i https://pypi.anaconda.org/numba/label/wheels_experimental_m1/simple numba
3. Remove pyaudio from the requirements file and instead execute this command to install pyaudio:

python -m pip install --global-option='build_ext' --global-option='-I/opt/homebrew/Cellar/portaudio/19.7.0/include' --global-option='-L/opt/homebrew/Cellar/portaudio/19.7.0/lib' pyaudio

4. Proceed with installing requirements as described above
5. Switch to the adviser folder cd adviser (containing the run_chat.py file)
6. Copy the snd library into the current folder:

cp /opt/homebrew/lib/libsndfile.dylib

Experimental Windows Instructions

NOTE: Windows support is not thoroughly tested so far and in experimental stage! Only tested on Windows 11 so far. If you encounter an error message about failing to build some library while installing the python dependencies, try installing the vcc build tools and repeat the failing step (https://visualstudio.microsoft.com/de/visual-cpp-build-tools/, yselect Desktop Development with C++ in installer).

0. Install Anaconda from https://www.anaconda.com/ IMPORTANT: The following commands have to be executed from the Anaconda prompt!
1. Create a virtual env for python3.8 using conda (there are no precompiled pyaudio packages for newer python versions at the time of writing)

conda create -n YOURVIRTUALENV python=3.8

2. Install pytorch from https://pytorch.org/get-started/locally/ . Select options build: stable, os0: windows, package: conda, language: python, compute platform: cuda XX.X if you have an NVIDIA GPU, else platform: cpu
3. Download sqlite3 precompiled library for Windows from https://www.sqlite.org/download.html . After unzipping, you will find a file sqlite3.dll. Copy this file into the DLL folder of your virtual environment (usually located at C:\Users\YOURSELF\anaconda3\envs\YOURVIRTUALENV\DLLs\).
4. Download and install grapviz installer for windows (version 4.X): https://www.graphviz.org/download/

If you don't want a multimodal setup, SKIP STEPS 5) and 6)

5. Install precompiled pyaudio

conda install pyaudio

6. Download trained models from http://adviserresources.ims.uni-stuttgart.de/models/adviser_models.zip and unzip into adviser/resources/models (folder 'models' does not exist initially)
7. Remove from the files requirements.txt and requirements_multimodal.txt the lines starting with torch, torchaudio, PyAudio.
8. Install the requirements from either `requirements.txt or requirements_multimodal.txt if you want a multimodal setup.

Building the documentation

1. Install the Python packages from requirements_doc.txt.
2. Run PYTHONPATH=./adviser mkdocs build or PYTHONPATH=./adviser mkdocs gh-deploy for pushing directly to GitHub Pages.

Support

You can ask questions by sending emails to [e-mail].

You can also post bug reports and feature requests in GitHub issues.

How to cite

If you use or reimplement any of this source code, please cite the following paper:

@InProceedings{
 title =     {ADVISER: A Toolkit for Developing Multi-modal, Multi-domain and Socially-engaged Conversational Agents},
 author =    {Chia-Yu Li and Daniel Ortega and Dirk V{\"{a}}th and Florian Lux and Lindsey Vanderlyn and Maximilian Schmidt and Michael Neumann and Moritz V{\"{o}}lkel and Pavel Denisov and Sabrina Jenne and Zorica Karacevic and Ngoc Thang Vu},
 booktitle = {Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020) - System Demonstrations},
 publisher = {Association for Computational Linguistics},
 location =  {Seattle, Washington, USA},
 year =      {2020}
 }

License

Adviser is published under the GNU GPL 3 license.