Replace tts google.py for home assistant with code that uses wavenet

google.py

"""
Support for the google speech service.

For more details about this component, please refer to the documentation at
https://home-assistant.io/components/tts.google/

Note - this is a hack. It makes no attempt to update tests. It does not have all wavenet voices listed.
It attempts to respect language requests from hass but this has not been tested. 

Google cloud gives the first 1 million characters of wavenet generation for free per month, if you exceed
that number they meter and charge. 

For this to work, you must have downloaded a provisioning key from google as detailed in this quickstart:
https://cloud.google.com/text-to-speech/docs/quickstart-client-libraries

Follow the instructions there, and make sure that you have the GOOGLE_APPLICATION_CREDENTIALS environment variable
set properly before booting hass.

"""
import asyncio
import logging
import re

import async_timeout
import voluptuous as vol
import yarl

from homeassistant.components.tts import CONF_LANG, PLATFORM_SCHEMA, Provider
from homeassistant.helpers.aiohttp_client import async_get_clientsession

from google.cloud import texttospeech

REQUIREMENTS = ['gTTS-token==1.1.3']

_LOGGER = logging.getLogger(__name__)

GOOGLE_SPEECH_URL = "https://translate.google.com/translate_tts"
MESSAGE_SIZE = 148

SUPPORT_LANGUAGES = [
    'af', 'sq', 'ar', 'hy', 'bn', 'ca', 'zh', 'zh-cn', 'zh-tw', 'zh-yue',
    'hr', 'cs', 'da', 'nl', 'en', 'en-au', 'en-uk', 'en-us', 'eo', 'fi',
    'fr', 'de', 'el', 'hi', 'hu', 'is', 'id', 'it', 'ja', 'ko', 'la', 'lv',
    'mk', 'no', 'pl', 'pt', 'pt-br', 'ro', 'ru', 'sr', 'sk', 'es', 'es-es',
    'es-mx', 'es-us', 'sw', 'sv', 'ta', 'th', 'tr', 'vi', 'cy', 'uk', 'bg-BG'
]

DEFAULT_LANG = 'fr'

WAVENET_LOOKUP = {'en': 'en-US-Wavenet-C', 'en-au': 'en-AU-Wavenet-A',
        'en-uk': 'en-GB-Wavenet-A', 'fr': 'fr-FR-Wavenet-A', 'de': 'de-DE-Wavenet-A',
        'it': 'it-IT-Wavenet-A', 'sv': 'sv-SE-Wavenet-A'}

PLATFORM_SCHEMA = PLATFORM_SCHEMA.extend({
    vol.Optional(CONF_LANG, default=DEFAULT_LANG): vol.In(SUPPORT_LANGUAGES),
})


async def async_get_engine(hass, config):
    """Set up Google speech component."""
    return GoogleProvider(hass, config[CONF_LANG])


class GoogleProvider(Provider):
    """The Google speech API provider."""

    def __init__(self, hass, lang):
        """Init Google TTS service."""
        self.hass = hass
        self._lang = lang
        self.name = 'Google'
        self.client = texttospeech.TextToSpeechClient()
        if self._lang in WAVENET_LOOKUP:
            language_code = WAVENET_LOOKUP[self._lang]
        else:
            language_code = WAVENET_LOOKUP['fr']

        self.voice = texttospeech.types.VoiceSelectionParams(
            language_code=language_code)

        self.audio_config = texttospeech.types.AudioConfig(
            audio_encoding=texttospeech.enums.AudioEncoding.MP3)


    @property
    def default_language(self):
        """Return the default language."""
        return self._lang

    @property
    def supported_languages(self):
        """Return list of supported languages."""
        return SUPPORT_LANGUAGES

    async def async_get_tts_audio(self, message, language, options=None):
        """Load TTS from google."""

        synthesis_input = texttospeech.types.SynthesisInput(text=message)

        try:
            with async_timeout.timeout(10, loop=self.hass.loop):
                response = self.client.synthesize_speech(
                    synthesis_input,
                    self.voice,
                    self.audio_config)

                return ("mp3", response.audio_content)

        except Exception as e:
            _LOGGER.error("Timeout for google speech. Or some other problem.", e)
            return (None, None)

    @staticmethod
    def _split_message_to_parts(message):
        """Split message into single parts."""
        if len(message) <= MESSAGE_SIZE:
            return [message]

        punc = "!()[]?.,;:"
        punc_list = [re.escape(c) for c in punc]
        pattern = '|'.join(punc_list)
        parts = re.split(pattern, message)

        def split_by_space(fullstring):
            """Split a string by space."""
            if len(fullstring) > MESSAGE_SIZE:
                idx = fullstring.rfind(' ', 0, MESSAGE_SIZE)
                return [fullstring[:idx]] + split_by_space(fullstring[idx:])
            return [fullstring]

        msg_parts = []
        for part in parts:
            msg_parts += split_by_space(part)

        return [msg for msg in msg_parts if len(msg) > 0]

This works for me (with hardcoded app credentials) and is the best tts so far, so thanks!

One minor issue, I can't get it to change the voice. I'm trying to switch to "en-US-Wavenet-F" but it sounds like "en-US-Wavenet-D" is the default. I can see it's pinging the app console, and switching between languages works too, it's just not recognizing the name specified in WAVENET_LOOKUP.

Any hints?

It's only important because Wavenet-F sounds more like Janet 😄