Voice Assistant

Voice assistant is a Python library to write a voice assistant.

This library is just an abstraction to build easily a voice assistant. It uses Google voice API so in this moments it only works with an Internet connection. Also remember to have a microphone and speakers or headphones connected to the computer before using it.

Basic assistant example ("hello world"):

from voice_assistant import Assistant

assistant = Assistant()

while True:
    assistant.speak('What do you need?')
    assistant.listen()

    if assistant.has_heard(['hello', 'hey', 'hi', good morning']):
        assistant.speak("Hello!, I'm ready for what you need.")

Requirements and installation:

Ubuntu/Debian users:

Before installing the Python package it's necessary to install system dependencies. Try typing:

sudo apt-get install python3 python3-pip python3-pyaudio portaudio19-dev sox libsox-fmt-mp3 python-dev build-essential swig libpulse-dev

Then, just type:

pip3 install -U voice_assistant

If you get an error trying to install the package, maybe you can solve it installing some of this system dependencies:

libffi-dev libxml2-dev libxslt1-dev libpq-dev libldap2-dev libsasl2-dev libssl-dev zlib1g-dev

Try to install them with sudo apt-get install <dependece> and then try to install the package again with: pip install voice_assistant.

How to use

Creating assistant

Just type:

from voice_assistant import Assistant

assistant = Assistant()

Default language is English. To use with another language you can instantiate it in this way:

assistant = Assistant(language='es')

Note: Don't worry if you see messages in the standard output when you instantiate an assistant. They won't be in future versions.

Speaking

Now you can make it to speak:

assistant.speak('hello')

You should listen "hello".

If you need to make it speak in another language just add a language parameter:

assistant.speak('¿Cómo estas?', language='es')

Listening

You can make the assistant to listen. To do it just type:

assistant.listen()

Assistant will be listening and waiting to hear something. Of course, it's necessary to have a working microphone connected to the computer. When you stop speaking assistant stop listening and prints in the terminal what it thinks you have said.

Of course is not the same trying to understand something if you expect to listen something in another language. So take in mind that by default assistant is trying to listen something in the same language that it speaks. However, you can also make it listen in another language just using the same approach than when you change its spoken language:

assistant.listen(language='es')

Note: When an assistant is initialized it listen ambient noise to know how loud it should listen, so if ambient noise changes a lot maybe you should initialize it again or use adjust_for_ambient_noise() method.

You can get the last listened string calling last_recognised property:

assistant.last_recognised
'hello'

But the most powerful way to use it is just typing:

assistant.has_heard('hello')
True

This method returns True or False depending if word is in last recognised sentence or not. It's not case sensitive so you can write words starting with capital letters or not.

The method has_heard can accept a word, but also accepts a list of words. If one of provided words is in the last recognised sentece, it will return True:

assistant.has_heard(['goodbye', 'kiss', 'hello'])
True

Sentences log

Assistant have a log with all listened sentences (in this moments it creates a simple SQLite database in the directory where is used).

To print all listened sentences just type:

assistant.get_all_listened_sentences()
['hello', 'how are you']

If you want to make it forget all listened words you can do it using:

assistant.forget_listened_sentences()

Now get_all_listened_sentences() should return an empty list.

The last_recognised property doesn't access to memory, it just show what it's in RAM. To get last recognised from database, you can use the get_last_recognised_from_memory() method.

Memory

Sometimes it could be interesting to make the assistant to memorize a specific thing. You can do it using memorize(keyword, value) method, providing a keyword and a value for that keyword:

assistant.memorize('name', 'Megara')

Now if you want to retrieve that value you can use remember(keyword) method:

assistant.remember('name')
['Megara']

Notice that result is a list of strings and not just a string. That's because it can memorize lots of different 'concepts' with the same 'keyword'.

assistant.memorize('name', 'Pilar')

assistant.remember('name')
['Megara', 'Pilar']

To retrieve just the last record you can use remember_last(keyword):

assistant.remember_last('name')
'Pilar'

Take in mind that if the keyword provided doesn't exists remember(keyword) will return an empty list and remember_last(word) an empty string.

Note: All thing are registered as strings. So if you want to remember a number and then use it, don't forget to cast it. Example: years = int(a.remember_last('years')).

Multiple assistants

There is no problem to create multiple assistants and use them at the same time, just remember that they probably won't be able to use the microphone and speakers at the same moment.

However, it's interesting to maintain they "memory" separated to discern between what an assistant has listened and what have the other listened or to separate what one knows and what the other knows. To achieve this you can just initialize them with different database_names:

from voice_assistant import Assistant

assistant = Assistant()

second_assistant = Assistant(database_name='second_memory')

Default database name is 'memory', so don't repeat it or specify a different database name for each assistant.

It is also possible to use the same database and just change an specific table name.

To do it you can type:

another_assistant = Assistant(memory_table='new_memory', listen_log_table='new_listen_log')

Default memory_table value 'memory' and default listen_log_table values is 'listen_log'.

Note: In this moment, when there is a Database error, it raises "DatabaseException". You can manage them importing it in this way: from voice_assistant import DatabaseException.