core/tests/components/tts/test_init.py

547 lines
19 KiB
Python

"""The tests for the TTS component."""
import ctypes
import os
import shutil
from unittest.mock import patch, PropertyMock
import requests
import homeassistant.components.http as http
import homeassistant.components.tts as tts
from homeassistant.components.tts.demo import DemoProvider
from homeassistant.components.media_player import (
SERVICE_PLAY_MEDIA, MEDIA_TYPE_MUSIC, ATTR_MEDIA_CONTENT_ID,
ATTR_MEDIA_CONTENT_TYPE, DOMAIN as DOMAIN_MP)
from homeassistant.bootstrap import setup_component
from tests.common import (
get_test_home_assistant, get_test_instance_port, assert_setup_component,
mock_service)
class TestTTS(object):
"""Test the Google speech component."""
def setup_method(self):
"""Setup things to be run when tests are started."""
self.hass = get_test_home_assistant()
self.demo_provider = DemoProvider('en')
self.default_tts_cache = self.hass.config.path(tts.DEFAULT_CACHE_DIR)
setup_component(
self.hass, http.DOMAIN,
{http.DOMAIN: {http.CONF_SERVER_PORT: get_test_instance_port()}})
def teardown_method(self):
"""Stop everything that was started."""
if os.path.isdir(self.default_tts_cache):
shutil.rmtree(self.default_tts_cache)
self.hass.stop()
def test_setup_component_demo(self):
"""Setup the demo platform with defaults."""
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
assert self.hass.services.has_service(tts.DOMAIN, 'demo_say')
assert self.hass.services.has_service(tts.DOMAIN, 'clear_cache')
@patch('os.mkdir', side_effect=OSError(2, "No access"))
def test_setup_component_demo_no_access_cache_folder(self, mock_mkdir):
"""Setup the demo platform with defaults."""
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
assert not setup_component(self.hass, tts.DOMAIN, config)
assert not self.hass.services.has_service(tts.DOMAIN, 'demo_say')
assert not self.hass.services.has_service(tts.DOMAIN, 'clear_cache')
def test_setup_component_and_test_service(self):
"""Setup the demo platform and call service."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
assert calls[0].data[ATTR_MEDIA_CONTENT_TYPE] == MEDIA_TYPE_MUSIC
assert calls[0].data[ATTR_MEDIA_CONTENT_ID].find(
"/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_en_-_demo.mp3") \
!= -1
assert os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3"))
def test_setup_component_and_test_service_with_config_language(self):
"""Setup the demo platform and call service."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
'language': 'de'
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
assert calls[0].data[ATTR_MEDIA_CONTENT_TYPE] == MEDIA_TYPE_MUSIC
assert calls[0].data[ATTR_MEDIA_CONTENT_ID].find(
"/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_de_-_demo.mp3") \
!= -1
assert os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_de_-_demo.mp3"))
def test_setup_component_and_test_service_with_wrong_conf_language(self):
"""Setup the demo platform and call service with wrong config."""
config = {
tts.DOMAIN: {
'platform': 'demo',
'language': 'ru'
}
}
with assert_setup_component(0, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
def test_setup_component_and_test_service_with_service_language(self):
"""Setup the demo platform and call service."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
tts.ATTR_LANGUAGE: "de",
})
self.hass.block_till_done()
assert len(calls) == 1
assert calls[0].data[ATTR_MEDIA_CONTENT_TYPE] == MEDIA_TYPE_MUSIC
assert calls[0].data[ATTR_MEDIA_CONTENT_ID].find(
"/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_de_-_demo.mp3") \
!= -1
assert os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_de_-_demo.mp3"))
def test_setup_component_test_service_with_wrong_service_language(self):
"""Setup the demo platform and call service."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
tts.ATTR_LANGUAGE: "lang",
})
self.hass.block_till_done()
assert len(calls) == 0
assert not os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_lang_-_demo.mp3"))
def test_setup_component_and_test_service_with_service_options(self):
"""Setup the demo platform and call service with options."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
tts.ATTR_LANGUAGE: "de",
tts.ATTR_OPTIONS: {
'voice': 'alex'
}
})
self.hass.block_till_done()
opt_hash = ctypes.c_size_t(hash(frozenset({'voice': 'alex'}))).value
assert len(calls) == 1
assert calls[0].data[ATTR_MEDIA_CONTENT_TYPE] == MEDIA_TYPE_MUSIC
assert calls[0].data[ATTR_MEDIA_CONTENT_ID].find(
"/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_de_{0}_demo.mp3".format(opt_hash)) \
!= -1
assert os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_de_{0}_demo.mp3".format(
opt_hash)))
@patch('homeassistant.components.tts.demo.DemoProvider.default_options',
new_callable=PropertyMock(return_value={'voice': 'alex'}))
def test_setup_component_and_test_with_service_options_def(self, def_mock):
"""Setup the demo platform and call service with default options."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
tts.ATTR_LANGUAGE: "de",
})
self.hass.block_till_done()
opt_hash = ctypes.c_size_t(hash(frozenset({'voice': 'alex'}))).value
assert len(calls) == 1
assert calls[0].data[ATTR_MEDIA_CONTENT_TYPE] == MEDIA_TYPE_MUSIC
assert calls[0].data[ATTR_MEDIA_CONTENT_ID].find(
"/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_de_{0}_demo.mp3".format(opt_hash)) \
!= -1
assert os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_de_{0}_demo.mp3".format(
opt_hash)))
def test_setup_component_and_test_service_with_service_options_wrong(self):
"""Setup the demo platform and call service with wrong options."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
tts.ATTR_LANGUAGE: "de",
tts.ATTR_OPTIONS: {
'speed': 1
}
})
self.hass.block_till_done()
opt_hash = ctypes.c_size_t(hash(frozenset({'speed': 1}))).value
assert len(calls) == 0
assert not os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_de_{0}_demo.mp3".format(
opt_hash)))
def test_setup_component_and_test_service_clear_cache(self):
"""Setup the demo platform and call service clear cache."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
assert os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3"))
self.hass.services.call(tts.DOMAIN, tts.SERVICE_CLEAR_CACHE, {})
self.hass.block_till_done()
assert not os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3"))
def test_setup_component_and_test_service_with_receive_voice(self):
"""Setup the demo platform and call service and receive voice."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.start()
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
req = requests.get(calls[0].data[ATTR_MEDIA_CONTENT_ID])
_, demo_data = self.demo_provider.get_tts_audio("bla", 'en')
assert req.status_code == 200
assert req.content == demo_data
def test_setup_component_and_test_service_with_receive_voice_german(self):
"""Setup the demo platform and call service and receive voice."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.start()
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
req = requests.get(calls[0].data[ATTR_MEDIA_CONTENT_ID])
_, demo_data = self.demo_provider.get_tts_audio("bla", "de")
assert req.status_code == 200
assert req.content == demo_data
def test_setup_component_and_web_view_wrong_file(self):
"""Setup the demo platform and receive wrong file from web."""
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.start()
url = ("{}/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_en_-_demo.mp3").format(self.hass.config.api.base_url)
req = requests.get(url)
assert req.status_code == 404
def test_setup_component_and_web_view_wrong_filename(self):
"""Setup the demo platform and receive wrong filename from web."""
config = {
tts.DOMAIN: {
'platform': 'demo',
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.start()
url = ("{}/api/tts_proxy/265944dsk32c1b2a621be5930510bb2cd"
"_en_-_demo.mp3").format(self.hass.config.api.base_url)
req = requests.get(url)
assert req.status_code == 404
def test_setup_component_test_without_cache(self):
"""Setup demo platform without cache."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
'cache': False,
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
assert not os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3"))
def test_setup_component_test_with_cache_call_service_without_cache(self):
"""Setup demo platform with cache and call service without cache."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo',
'cache': True,
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
tts.ATTR_CACHE: False,
})
self.hass.block_till_done()
assert len(calls) == 1
assert not os.path.isfile(os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3"))
def test_setup_component_test_with_cache_dir(self):
"""Setup demo platform with cache and call service without cache."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
_, demo_data = self.demo_provider.get_tts_audio("bla", 'en')
cache_file = os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3")
os.mkdir(self.default_tts_cache)
with open(cache_file, "wb") as voice_file:
voice_file.write(demo_data)
config = {
tts.DOMAIN: {
'platform': 'demo',
'cache': True,
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
with patch('homeassistant.components.tts.demo.DemoProvider.'
'get_tts_audio', return_value=(None, None)):
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 1
assert calls[0].data[ATTR_MEDIA_CONTENT_ID].find(
"/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_en_-_demo.mp3") \
!= -1
@patch('homeassistant.components.tts.demo.DemoProvider.get_tts_audio',
return_value=(None, None))
def test_setup_component_test_with_error_on_get_tts(self, tts_mock):
"""Setup demo platform with wrong get_tts_audio."""
calls = mock_service(self.hass, DOMAIN_MP, SERVICE_PLAY_MEDIA)
config = {
tts.DOMAIN: {
'platform': 'demo'
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.services.call(tts.DOMAIN, 'demo_say', {
tts.ATTR_MESSAGE: "I person is on front of your door.",
})
self.hass.block_till_done()
assert len(calls) == 0
def test_setup_component_load_cache_retrieve_without_mem_cache(self):
"""Setup component and load cache and get without mem cache."""
_, demo_data = self.demo_provider.get_tts_audio("bla", 'en')
cache_file = os.path.join(
self.default_tts_cache,
"265944c108cbb00b2a621be5930513e03a0bb2cd_en_-_demo.mp3")
os.mkdir(self.default_tts_cache)
with open(cache_file, "wb") as voice_file:
voice_file.write(demo_data)
config = {
tts.DOMAIN: {
'platform': 'demo',
'cache': True,
}
}
with assert_setup_component(1, tts.DOMAIN):
setup_component(self.hass, tts.DOMAIN, config)
self.hass.start()
url = ("{}/api/tts_proxy/265944c108cbb00b2a621be5930513e03a0bb2cd"
"_en_-_demo.mp3").format(self.hass.config.api.base_url)
req = requests.get(url)
assert req.status_code == 200
assert req.content == demo_data