Alexa Voice Service Client

Python Client for Alexa Voice Service (AVS)

Installation

pip install alexa_client

or if you want to run the demos:

pip install alexa_client[demo]

Usage

File audio

from alexa_client import AlexaClient

client = AlexaClient(
    client_id='my-client-id',
    secret='my-secret',
    refresh_token='my-refresh-token',
)
client.connect()  # authenticate and other handshaking steps
with open('./tests/resources/alexa_what_time_is_it.wav', 'rb') as f:
    for i, directive in enumerate(client.send_audio_file(f)):
        if directive.name in ['Speak', 'Play']:
            with open(f'./output_{i}.mp3', 'wb') as f:
                f.write(directive.audio_attachment)

Now listen to output_0.wav and Alexa should tell you the time.

Microphone audio

import io

from alexa_client import AlexaClient
import pyaudio


def callback(in_data, frame_count, time_info, status):
    buffer.write(in_data)
    return (in_data, pyaudio.paContinue)

p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=1,
    rate=16000,
    input=True,
    stream_callback=callback,
)

client = AlexaClient(
    client_id='my-client-id',
    secret='my-secret',
    refresh_token='my-refresh-token',
)

buffer = io.BytesIO()
try:
    stream.start_stream()
    print('listening. Press CTRL + C to exit.')
    client.connect()
    for i, directive in enumerate(client.send_audio_file(buffer)):
        if directive.name in ['Speak', 'Play']:
            with open(f'./output_{i}.mp3', 'wb') as f:
                f.write(directive.audio_attachment)
finally:
    stream.stop_stream()
    stream.close()
    p.terminate()

Multi-step requests

An Alexa command may relate to a previous command e.g,

[you] "Alexa, play twenty questions" [Alexa] "Is it a animal, mineral, or vegetable?" [you] "Mineral" [Alexa] "Is it valuable" [you] "No" [Alexa] "is it..."

This can be achieved by passing the same dialog request ID to multiple send_audio_file calls:

from alexa_client.alexa_client import helpers

dialog_request_id = helpers.generate_unique_id()
directives_one = client.send_audio_file(audio_one, dialog_request_id=dialog_request_id)
directives_two = client.send_audio_file(audio_two, dialog_request_id=dialog_request_id)
directives_three = client.send_audio_file(audio_three, dialog_request_id=dialog_request_id)

Run the streaming microphone audio demo to use this feature:

pip install alexa_client[demo]
python -m alexa_client.demo.streaming_microphone \
    --client-id="{enter-client-id-here}" \
    --client-secret="{enter-client-secret-here}" \
    --refresh-token="{enter-refresh-token-here}"

ASR Profiles

Automatic Speech Recognition (ASR) profiles optimized for user speech from varying distances. By default CLOSE_TALK is used but this can be specified:

from alexa_client import constants

client.send_audio_file(
    audio_file=audio_file,
    distance_profile=constants.NEAR_FIELD,  # or constants.FAR_FIELD
)

Audio format

By default PCM audio format is assumed, but OPUS can be specified:

from alexa_client import constants

client.send_audio_file(
    audio_file=audio_file,
    audio_format=constants.OPUS,
)

When PCM format is specified the audio should be 16bit Linear PCM (LPCM16), 16kHz sample rate, single-channel, and little endian.

When OPUS forat is specified the audio should be 16bit Opus, 16kHz sample rate, 32k bit rate, and little endian.

Base URL

base_url can be set to improve latency. Choose a region closest to your location.

from alexa_client.alexa_client import constants

client = AlexaClient(
    client_id='my-client-id',
    secret='my-secret',
    refresh_token='my-refresh-token',
    base_url=constants.BASE_URL_ASIA
)

The default base URL is Europe. The available constants are BASE_URL_EUROPE, BASE_URL_ASIA and BASE_URL_NORTH_AMERICA but you can pass any string if required.

Authentication

To use AVS you must first have a developer account. Then register your product here. Choose "Application" under "Is your product an app or a device"?

The client requires your client_id, secret and refresh_token:

client kwarg	Notes
`client_id`	Retrieve by clicking on the your product listed here
`secret`	Retrieve by clicking on the your product listed here
`refresh_token`	You must generate this. See below

Refresh token

You will need to login to Amazon via a web browser to get your refresh token.

To enable this first go here and click on your product to set some security settings under Security Profile:

setting	value
Allowed Origins	http://localhost:9000
Allowed Return URLs	http://localhost:9000/callback/

Note what you entered for Product ID under Product Information, as this will be used as the device-type-id (case sensitive!)

Then run:

python -m alexa_client.refreshtoken.serve \
    --device-type-id="{enter-device-type-id-here}" \
    --client-id="{enter-client-id-here}" \
    --client-secret="{enter-client-secret-here}"

Follow the on-screen instructions shown at http://localhost:9000 in your web browser. On completion Amazon will return your refresh_token - which you will require to send audio or recorded voice.

Steaming audio to AVS

AlexaClient.send_audio_file streaming uploads a file-like object to AVS for great latency. The file-like object can be an actual file on your filesystem, an in-memory BytesIo buffer containing audio from your microphone, or even audio streaming from your browser over a websocket in real-time.

Persistent AVS connection

Calling AlexaClient.connect creates a persistent connection to AVS. A thread runs that pings AVS after 4 minutes of no request being made to AVS. This prevents the connection getting forcefully closed due to inactivity.

Unit test

To run the unit tests, call the following commands:

git clone git@github.com:richtier/alexa-voice-service-client.git
make test_requirements
pytest

Other projects

This library is used by alexa-browser-client, which allows you to talk to Alexa from your browser.

rdematos/alexa-voice-service-client