git.basschouten.com Git - openhab-addons.git/blob

   1 /**
   2  * Copyright (c) 2010-2023 Contributors to the openHAB project
   3  *
   4  * See the NOTICE file(s) distributed with this work for additional
   5  * information.
   6  *
   7  * This program and the accompanying materials are made available under the
   8  * terms of the Eclipse Public License 2.0 which is available at
   9  * http://www.eclipse.org/legal/epl-2.0
  10  *
  11  * SPDX-License-Identifier: EPL-2.0
  12  */
  13 package org.openhab.voice.googletts.internal;
  14
  15 import java.io.IOException;
  16 import java.nio.charset.StandardCharsets;
  17 import java.util.ArrayList;
  18 import java.util.Base64;
  19 import java.util.Dictionary;
  20 import java.util.HashMap;
  21 import java.util.HashSet;
  22 import java.util.List;
  23 import java.util.Locale;
  24 import java.util.Map;
  25 import java.util.Set;
  26
  27 import org.eclipse.jdt.annotation.Nullable;
  28 import org.eclipse.jetty.http.HttpHeader;
  29 import org.eclipse.jetty.http.MimeTypes;
  30 import org.openhab.core.audio.AudioFormat;
  31 import org.openhab.core.auth.AuthenticationException;
  32 import org.openhab.core.auth.client.oauth2.AccessTokenResponse;
  33 import org.openhab.core.auth.client.oauth2.OAuthClientService;
  34 import org.openhab.core.auth.client.oauth2.OAuthException;
  35 import org.openhab.core.auth.client.oauth2.OAuthFactory;
  36 import org.openhab.core.auth.client.oauth2.OAuthResponseException;
  37 import org.openhab.core.i18n.CommunicationException;
  38 import org.openhab.core.io.net.http.HttpRequestBuilder;
  39 import org.openhab.voice.googletts.internal.dto.AudioConfig;
  40 import org.openhab.voice.googletts.internal.dto.AudioEncoding;
  41 import org.openhab.voice.googletts.internal.dto.ListVoicesResponse;
  42 import org.openhab.voice.googletts.internal.dto.SsmlVoiceGender;
  43 import org.openhab.voice.googletts.internal.dto.SynthesisInput;
  44 import org.openhab.voice.googletts.internal.dto.SynthesizeSpeechRequest;
  45 import org.openhab.voice.googletts.internal.dto.SynthesizeSpeechResponse;
  46 import org.openhab.voice.googletts.internal.dto.Voice;
  47 import org.openhab.voice.googletts.internal.dto.VoiceSelectionParams;
  48 import org.osgi.service.cm.Configuration;
  49 import org.osgi.service.cm.ConfigurationAdmin;
  50 import org.slf4j.Logger;
  51 import org.slf4j.LoggerFactory;
  52
  53 import com.google.gson.Gson;
  54 import com.google.gson.GsonBuilder;
  55 import com.google.gson.JsonSyntaxException;
  56
  57 /**
  58  * Google Cloud TTS API call implementation.
  59  *
  60  * @author Gabor Bicskei - Initial contribution and API
  61  */
  62 class GoogleCloudAPI {
  63
  64     private static final String BEARER = "Bearer ";
  65
  66     private static final String GCP_AUTH_URI = "https://accounts.google.com/o/oauth2/auth";
  67     private static final String GCP_TOKEN_URI = "https://accounts.google.com/o/oauth2/token";
  68     private static final String GCP_REDIRECT_URI = "https://www.google.com";
  69     /**
  70      * Google Cloud Platform authorization scope
  71      */
  72     private static final String GCP_SCOPE = "https://www.googleapis.com/auth/cloud-platform";
  73
  74     /**
  75      * URL used for retrieving the list of available voices
  76      */
  77     private static final String LIST_VOICES_URL = "https://texttospeech.googleapis.com/v1/voices";
  78
  79     /**
  80      * URL used for synthesizing text to speech
  81      */
  82     private static final String SYTNHESIZE_SPEECH_URL = "https://texttospeech.googleapis.com/v1/text:synthesize";
  83
  84     /**
  85      * Logger
  86      */
  87     private final Logger logger = LoggerFactory.getLogger(GoogleCloudAPI.class);
  88
  89     /**
  90      * Supported voices and locales
  91      */
  92     private final Map<Locale, Set<GoogleTTSVoice>> voices = new HashMap<>();
  93
  94     /**
  95      * Configuration
  96      */
  97     private @Nullable GoogleTTSConfig config;
  98
  99     private final Gson gson = new GsonBuilder().create();
 100     private final ConfigurationAdmin configAdmin;
 101     private final OAuthFactory oAuthFactory;
 102
 103     private @Nullable OAuthClientService oAuthService;
 104
 105     /**
 106      * Constructor.
 107      *
 108      */
 109     GoogleCloudAPI(ConfigurationAdmin configAdmin, OAuthFactory oAuthFactory) {
 110         this.configAdmin = configAdmin;
 111         this.oAuthFactory = oAuthFactory;
 112     }
 113
 114     /**
 115      * Configuration update.
 116      *
 117      * @param config New configuration.
 118      */
 119     void setConfig(GoogleTTSConfig config) {
 120         this.config = config;
 121
 122         if (oAuthService != null) {
 123             oAuthFactory.ungetOAuthService(GoogleTTSService.SERVICE_PID);
 124             oAuthService = null;
 125         }
 126
 127         String clientId = config.clientId;
 128         String clientSecret = config.clientSecret;
 129         if (clientId != null && !clientId.isEmpty() && clientSecret != null && !clientSecret.isEmpty()) {
 130             final OAuthClientService oAuthService = oAuthFactory.createOAuthClientService(GoogleTTSService.SERVICE_PID,
 131                     GCP_TOKEN_URI, GCP_AUTH_URI, clientId, clientSecret, GCP_SCOPE, false);
 132             this.oAuthService = oAuthService;
 133             try {
 134                 getAccessToken();
 135                 initVoices();
 136             } catch (AuthenticationException | CommunicationException e) {
 137                 logger.warn("Error initializing Google Cloud TTS service: {}", e.getMessage());
 138                 oAuthFactory.ungetOAuthService(GoogleTTSService.SERVICE_PID);
 139                 this.oAuthService = null;
 140                 voices.clear();
 141             }
 142         } else {
 143             voices.clear();
 144         }
 145     }
 146
 147     public void dispose() {
 148         if (oAuthService != null) {
 149             oAuthFactory.ungetOAuthService(GoogleTTSService.SERVICE_PID);
 150             oAuthService = null;
 151         }
 152         voices.clear();
 153     }
 154
 155     /**
 156      * Fetches the OAuth2 tokens from Google Cloud Platform if the auth-code is set in the configuration. If successful
 157      * the auth-code will be removed from the configuration.
 158      *
 159      * @throws AuthenticationException
 160      * @throws CommunicationException
 161      */
 162     @SuppressWarnings("null")
 163     private void getAccessToken() throws AuthenticationException, CommunicationException {
 164         String authcode = config.authcode;
 165         if (authcode != null && !authcode.isEmpty()) {
 166             logger.debug("Trying to get access and refresh tokens.");
 167             try {
 168                 AccessTokenResponse response = oAuthService.getAccessTokenResponseByAuthorizationCode(authcode,
 169                         GCP_REDIRECT_URI);
 170                 if (response.getRefreshToken() == null || response.getRefreshToken().isEmpty()) {
 171                     throw new AuthenticationException("Error fetching refresh token. Please reauthorize");
 172                 }
 173             } catch (OAuthException | OAuthResponseException e) {
 174                 logger.debug("Error fetching access token: {}", e.getMessage(), e);
 175                 throw new AuthenticationException(
 176                         "Error fetching access token. Invalid authcode? Please generate a new one.");
 177             } catch (IOException e) {
 178                 throw new CommunicationException(
 179                         String.format("An unexpected IOException occurred: %s", e.getMessage()));
 180             }
 181
 182             config.authcode = null;
 183
 184             try {
 185                 Configuration serviceConfig = configAdmin.getConfiguration(GoogleTTSService.SERVICE_PID);
 186                 Dictionary<String, Object> configProperties = serviceConfig.getProperties();
 187                 if (configProperties != null) {
 188                     configProperties.put(GoogleTTSService.PARAM_AUTHCODE, "");
 189                     serviceConfig.update(configProperties);
 190                 }
 191             } catch (IOException e) {
 192                 // should not happen
 193                 logger.warn(
 194                         "Failed to update configuration for Google Cloud TTS service. Please clear the 'authcode' configuration parameter manualy.");
 195             }
 196         }
 197     }
 198
 199     @SuppressWarnings("null")
 200     private String getAuthorizationHeader() throws AuthenticationException, CommunicationException {
 201         final AccessTokenResponse accessTokenResponse;
 202         try {
 203             accessTokenResponse = oAuthService.getAccessTokenResponse();
 204         } catch (OAuthException | OAuthResponseException e) {
 205             logger.debug("Error fetching access token: {}", e.getMessage(), e);
 206             throw new AuthenticationException(
 207                     "Error fetching access token. Invalid authcode? Please generate a new one.");
 208         } catch (IOException e) {
 209             throw new CommunicationException(String.format("An unexpected IOException occurred: %s", e.getMessage()));
 210         }
 211         if (accessTokenResponse == null || accessTokenResponse.getAccessToken() == null
 212                 || accessTokenResponse.getAccessToken().isEmpty()) {
 213             throw new AuthenticationException("No access token. Is this thing authorized?");
 214         }
 215         if (accessTokenResponse.getRefreshToken() == null || accessTokenResponse.getRefreshToken().isEmpty()) {
 216             throw new AuthenticationException("No refresh token. Please reauthorize");
 217         }
 218         return BEARER + accessTokenResponse.getAccessToken();
 219     }
 220
 221     /**
 222      * Loads supported audio formats
 223      *
 224      * @return Set of audio formats
 225      */
 226     Set<String> getSupportedAudioFormats() {
 227         Set<String> formats = new HashSet<>();
 228         for (AudioEncoding audioEncoding : AudioEncoding.values()) {
 229             if (audioEncoding != AudioEncoding.AUDIO_ENCODING_UNSPECIFIED) {
 230                 formats.add(audioEncoding.toString());
 231             }
 232         }
 233         return formats;
 234     }
 235
 236     /**
 237      * Supported locales.
 238      *
 239      * @return Set of locales
 240      */
 241     Set<Locale> getSupportedLocales() {
 242         return voices.keySet();
 243     }
 244
 245     /**
 246      * Supported voices for locale.
 247      *
 248      * @param locale Locale
 249      * @return Set of voices
 250      */
 251     Set<GoogleTTSVoice> getVoicesForLocale(Locale locale) {
 252         Set<GoogleTTSVoice> localeVoices = voices.get(locale);
 253         return localeVoices != null ? localeVoices : Set.of();
 254     }
 255
 256     /**
 257      * Google API call to load locales and voices.
 258      *
 259      * @throws AuthenticationException
 260      * @throws CommunicationException
 261      */
 262     private void initVoices() throws AuthenticationException, CommunicationException {
 263         if (oAuthService != null) {
 264             voices.clear();
 265             for (GoogleTTSVoice voice : listVoices()) {
 266                 Locale locale = voice.getLocale();
 267                 Set<GoogleTTSVoice> localeVoices;
 268                 if (!voices.containsKey(locale)) {
 269                     localeVoices = new HashSet<>();
 270                     voices.put(locale, localeVoices);
 271                 } else {
 272                     localeVoices = voices.get(locale);
 273                 }
 274                 localeVoices.add(voice);
 275             }
 276         } else {
 277             logger.error("Google client is not initialized!");
 278         }
 279     }
 280
 281     @SuppressWarnings("null")
 282     private List<GoogleTTSVoice> listVoices() throws AuthenticationException, CommunicationException {
 283         HttpRequestBuilder builder = HttpRequestBuilder.getFrom(LIST_VOICES_URL)
 284                 .withHeader(HttpHeader.AUTHORIZATION.name(), getAuthorizationHeader());
 285
 286         try {
 287             ListVoicesResponse listVoicesResponse = gson.fromJson(builder.getContentAsString(),
 288                     ListVoicesResponse.class);
 289
 290             if (listVoicesResponse == null || listVoicesResponse.getVoices() == null) {
 291                 return List.of();
 292             }
 293
 294             List<GoogleTTSVoice> result = new ArrayList<>();
 295             for (Voice voice : listVoicesResponse.getVoices()) {
 296                 for (String languageCode : voice.getLanguageCodes()) {
 297                     result.add(new GoogleTTSVoice(Locale.forLanguageTag(languageCode), voice.getName(),
 298                             voice.getSsmlGender().name()));
 299                 }
 300             }
 301             return result;
 302         } catch (JsonSyntaxException e) {
 303             // do nothing
 304         } catch (IOException e) {
 305             throw new CommunicationException(String.format("An unexpected IOException occurred: %s", e.getMessage()));
 306         }
 307         return List.of();
 308     }
 309
 310     /**
 311      * Converts audio format to Google parameters.
 312      *
 313      * @param codec Requested codec
 314      * @return String array of Google audio format and the file extension to use.
 315      */
 316     private String getFormatForCodec(String codec) {
 317         switch (codec) {
 318             case AudioFormat.CODEC_MP3:
 319                 return AudioEncoding.MP3.toString();
 320             case AudioFormat.CODEC_PCM_SIGNED:
 321                 return AudioEncoding.LINEAR16.toString();
 322             default:
 323                 throw new IllegalArgumentException("Audio format " + codec + " is not yet supported");
 324         }
 325     }
 326
 327     public byte[] synthesizeSpeech(String text, GoogleTTSVoice voice, String codec) {
 328         String format = getFormatForCodec(codec);
 329         try {
 330             return synthesizeSpeechByGoogle(text, voice, format);
 331         } catch (AuthenticationException | CommunicationException e) {
 332             logger.warn("Error initializing Google Cloud TTS service: {}", e.getMessage());
 333             if (oAuthService != null) {
 334                 oAuthFactory.ungetOAuthService(GoogleTTSService.SERVICE_PID);
 335                 oAuthService = null;
 336             }
 337             voices.clear();
 338         }
 339         return null;
 340     }
 341
 342     /**
 343      * Call Google service to synthesize the required text
 344      *
 345      * @param text Text to synthesize
 346      * @param voice Voice parameter
 347      * @param audioFormat Audio encoding format
 348      * @return Audio input stream or {@code null} when encoding exceptions occur
 349      * @throws AuthenticationException
 350      * @throws CommunicationException
 351      */
 352     @SuppressWarnings("null")
 353     private byte[] synthesizeSpeechByGoogle(String text, GoogleTTSVoice voice, String audioFormat)
 354             throws AuthenticationException, CommunicationException {
 355         AudioConfig audioConfig = new AudioConfig(AudioEncoding.valueOf(audioFormat), config.pitch, config.speakingRate,
 356                 config.volumeGainDb);
 357         SynthesisInput synthesisInput = new SynthesisInput(text);
 358         VoiceSelectionParams voiceSelectionParams = new VoiceSelectionParams(voice.getLocale().getLanguage(),
 359                 voice.getLabel(), SsmlVoiceGender.valueOf(voice.getSsmlGender()));
 360
 361         SynthesizeSpeechRequest request = new SynthesizeSpeechRequest(audioConfig, synthesisInput,
 362                 voiceSelectionParams);
 363
 364         HttpRequestBuilder builder = HttpRequestBuilder.postTo(SYTNHESIZE_SPEECH_URL)
 365                 .withHeader(HttpHeader.AUTHORIZATION.name(), getAuthorizationHeader())
 366                 .withContent(gson.toJson(request), MimeTypes.Type.APPLICATION_JSON.name());
 367
 368         try {
 369             SynthesizeSpeechResponse synthesizeSpeechResponse = gson.fromJson(builder.getContentAsString(),
 370                     SynthesizeSpeechResponse.class);
 371
 372             if (synthesizeSpeechResponse == null) {
 373                 return null;
 374             }
 375
 376             byte[] encodedBytes = synthesizeSpeechResponse.getAudioContent().getBytes(StandardCharsets.UTF_8);
 377             return Base64.getDecoder().decode(encodedBytes);
 378         } catch (JsonSyntaxException e) {
 379             // do nothing
 380         } catch (IOException e) {
 381             throw new CommunicationException(String.format("An unexpected IOException occurred: %s", e.getMessage()));
 382         }
 383         return null;
 384     }
 385
 386     boolean isInitialized() {
 387         return oAuthService != null;
 388     }
 389 }