|
| 1 | +/* |
| 2 | + * Copyright (c) 2024 by Naohide Sano, All rights reserved. |
| 3 | + * |
| 4 | + * Programmed by Naohide Sano |
| 5 | + */ |
| 6 | + |
| 7 | +package vavi.speech.coeiroink; |
| 8 | + |
| 9 | +import java.io.Closeable; |
| 10 | +import java.io.IOException; |
| 11 | +import java.io.InputStream; |
| 12 | +import java.util.ArrayList; |
| 13 | +import java.util.Arrays; |
| 14 | +import java.util.Collection; |
| 15 | +import java.util.List; |
| 16 | +import java.util.StringJoiner; |
| 17 | +import java.util.concurrent.Callable; |
| 18 | +import java.util.function.Function; |
| 19 | +import java.util.logging.Level; |
| 20 | +import java.util.logging.Logger; |
| 21 | + |
| 22 | +import com.google.api.client.util.ExponentialBackOff; |
| 23 | +import com.google.gson.Gson; |
| 24 | +import com.google.gson.GsonBuilder; |
| 25 | +import jakarta.ws.rs.client.Client; |
| 26 | +import jakarta.ws.rs.client.ClientBuilder; |
| 27 | +import jakarta.ws.rs.client.Entity; |
| 28 | +import jakarta.ws.rs.client.WebTarget; |
| 29 | +import jakarta.ws.rs.core.MediaType; |
| 30 | +import jakarta.ws.rs.core.Response; |
| 31 | +import jakarta.ws.rs.core.Response.Status.Family; |
| 32 | +import vavi.speech.voicevox.VoiceVox; |
| 33 | +import vavi.util.CharNormalizerJa; |
| 34 | +import vavi.util.Debug; |
| 35 | + |
| 36 | + |
| 37 | +/** |
| 38 | + * CoeiroInk. |
| 39 | + * |
| 40 | + * @author <a href="mailto:umjammer@gmail.com">Naohide Sano</a> (nsano) |
| 41 | + * @version 0.00 2024-04-02 nsano initial version <br> |
| 42 | + */ |
| 43 | +public class CoeiroInk implements Closeable { |
| 44 | + |
| 45 | + private static final Logger logger = Logger.getLogger(CoeiroInk.class.getName()); |
| 46 | + |
| 47 | + /** VoiceVox application web api */ |
| 48 | + private static String url = "http://127.0.0.1:50032/"; |
| 49 | + |
| 50 | + /** */ |
| 51 | + private static final Gson gson = new GsonBuilder().create(); |
| 52 | + |
| 53 | + /* */ |
| 54 | + static { |
| 55 | + String url = System.getProperty("vavi.speech.coeiroink.url"); |
| 56 | + if (url != null) { |
| 57 | + CoeiroInk.url = url; |
| 58 | + } |
| 59 | + } |
| 60 | + |
| 61 | + /** */ |
| 62 | + private final WebTarget target; |
| 63 | + |
| 64 | + /** */ |
| 65 | + private VoiceVox.Speaker[] speakers; |
| 66 | + |
| 67 | + /** */ |
| 68 | + private final Client client; |
| 69 | + |
| 70 | + /** */ |
| 71 | + public CoeiroInk() { |
| 72 | + try { |
| 73 | + client = ClientBuilder.newClient(); // DON'T CLOSE |
| 74 | + target = client.target(url); |
| 75 | + |
| 76 | + String json = target |
| 77 | + .path("/v1/engine_info") |
| 78 | + .request() |
| 79 | + .get(String.class); |
| 80 | + EngineInfo engineInfo = gson.fromJson(json, EngineInfo.class); |
| 81 | +Debug.println(Level.FINE, "version: " + engineInfo.version); |
| 82 | + } catch (Exception e) { |
| 83 | + throw new IllegalStateException("CoeiroInk is not available at " + url, e); |
| 84 | + } |
| 85 | + } |
| 86 | + |
| 87 | + @Override |
| 88 | + public void close() throws IOException { |
| 89 | + client.close(); |
| 90 | + } |
| 91 | + |
| 92 | + /** */ |
| 93 | + public Prosody getProsody(String text) { |
| 94 | + Entity<String> entity = Entity.entity("{ \"text\": \"" + text + "\"}", MediaType.APPLICATION_JSON); |
| 95 | + String json = target |
| 96 | + .path("v1/estimate_prosody") |
| 97 | + .request() |
| 98 | + .post(entity, String.class); |
| 99 | + return gson.fromJson(json, Prosody.class); |
| 100 | + } |
| 101 | + |
| 102 | + /** */ |
| 103 | + public InputStream synthesize(Synthesis synthesis) { |
| 104 | + Entity<String> syntheEntity = Entity.entity(gson.toJson(synthesis), MediaType.APPLICATION_JSON); |
| 105 | + return target.path("v1/synthesis") |
| 106 | + .request() |
| 107 | + .post(syntheEntity, InputStream.class); |
| 108 | + } |
| 109 | + |
| 110 | + /** */ |
| 111 | + public Speaker[] getAllVoices() { |
| 112 | + String speakersJson = target |
| 113 | + .path("/v1/speakers") |
| 114 | + .request() |
| 115 | + .get(String.class); |
| 116 | + return gson.fromJson(speakersJson, Speaker[].class); |
| 117 | + } |
| 118 | + |
| 119 | + //---- |
| 120 | + |
| 121 | + public static class EngineInfo { |
| 122 | + public String device; |
| 123 | + public String version; |
| 124 | + } |
| 125 | + |
| 126 | + public static class Speaker { |
| 127 | + public String speakerName; |
| 128 | + public String speakerUuid; |
| 129 | + public static class Style { |
| 130 | + public String styleName; |
| 131 | + public int styleId; |
| 132 | + public String base64Icon; |
| 133 | + public String base64Portrait; |
| 134 | + @Override public String toString() { |
| 135 | + return new StringJoiner(", ", Style.class.getSimpleName() + "[", "]") |
| 136 | + .add("styleName='" + styleName + "'") |
| 137 | + .add("styleId=" + styleId) |
| 138 | + .toString(); |
| 139 | + } |
| 140 | + } |
| 141 | + public Style[] styles; |
| 142 | + public String version; |
| 143 | + public String base64Portrait; |
| 144 | + @Override public String toString() { |
| 145 | + return new StringJoiner(", ", Speaker.class.getSimpleName() + "[", "]") |
| 146 | + .add("speakerName='" + speakerName + "'") |
| 147 | + .add("speakerUuid='" + speakerUuid + "'") |
| 148 | + .add("styles=" + Arrays.toString(styles)) |
| 149 | + .add("version='" + version + "'") |
| 150 | + .toString(); |
| 151 | + } |
| 152 | + } |
| 153 | + |
| 154 | + public static class Detail { |
| 155 | + public String phoneme; |
| 156 | + public String hira; |
| 157 | + public int accent; |
| 158 | + |
| 159 | + public Detail(String phoneme, String hira, int accent) { |
| 160 | + this.phoneme = phoneme; |
| 161 | + this.hira = hira; |
| 162 | + this.accent = accent; |
| 163 | + } |
| 164 | + } |
| 165 | + |
| 166 | + public static class Prosody { |
| 167 | + public String[] plain; |
| 168 | + public Detail[][] detail; |
| 169 | + |
| 170 | + @Override public String toString() { |
| 171 | + return new StringJoiner(", ", Prosody.class.getSimpleName() + "[", "]") |
| 172 | + .add("plain=" + Arrays.toString(plain)) |
| 173 | + .add("detail=" + Arrays.toString(detail)) |
| 174 | + .toString(); |
| 175 | + } |
| 176 | + } |
| 177 | + |
| 178 | + public static class Mora { |
| 179 | + public String text; |
| 180 | + public String consonant; |
| 181 | + public Integer consonant_length; |
| 182 | + public String vowel; |
| 183 | + public int vowel_length; |
| 184 | + public int pitch; |
| 185 | + |
| 186 | + public Mora(String text, String consonant, Integer consonant_length, String vowel, int vowel_length, int pitch) { |
| 187 | + this.text = text; |
| 188 | + this.consonant = consonant; |
| 189 | + this.consonant_length = consonant_length; |
| 190 | + this.vowel = vowel; |
| 191 | + this.vowel_length = vowel_length; |
| 192 | + this.pitch = pitch; |
| 193 | + } |
| 194 | + } |
| 195 | + |
| 196 | + public static class AccentPhrase { |
| 197 | + public Mora[] moras; |
| 198 | + public int accent; |
| 199 | + public boolean is_interrogative; |
| 200 | + public Mora pause_mora; |
| 201 | + |
| 202 | + public AccentPhrase(Mora[] moras, int accent, boolean is_interrogative, Mora pause_mora) { |
| 203 | + this.moras = moras; |
| 204 | + this.accent = accent; |
| 205 | + this.is_interrogative = is_interrogative; |
| 206 | + this.pause_mora = pause_mora; |
| 207 | + } |
| 208 | + } |
| 209 | + |
| 210 | + Function<Prosody, AccentPhrase[]> prosodyToAccentPhrases = prosody -> { |
| 211 | + List<AccentPhrase> result = new ArrayList<>(); |
| 212 | + for (var d : prosody.detail) { |
| 213 | + int accentPosition = -1; |
| 214 | + List<Mora> _moras = new ArrayList<>(); |
| 215 | + int moraIndex = -1; |
| 216 | + for (var m : d) { |
| 217 | + moraIndex++; |
| 218 | + if (m.hira.equals("、")) { |
| 219 | + result.get(result.size() - 1).pause_mora = new Mora( |
| 220 | + "、", |
| 221 | + null, |
| 222 | + null, |
| 223 | + "pau", |
| 224 | + 0, |
| 225 | + 0 |
| 226 | + ); |
| 227 | + } else { |
| 228 | + String _vowel, consonant; |
| 229 | + if (m.phoneme.contains("-")) { |
| 230 | + String[] pair = m.phoneme.split("-"); |
| 231 | + consonant = pair[0]; |
| 232 | + _vowel = pair[1]; |
| 233 | + } else { |
| 234 | + consonant = null; |
| 235 | + _vowel = m.phoneme; |
| 236 | + } |
| 237 | + _moras.add(new Mora( |
| 238 | + CharNormalizerJa.ToKatakana.normalize(m.hira), |
| 239 | + consonant, |
| 240 | + consonant != null ? 0 : null, |
| 241 | + _vowel, |
| 242 | + 0, |
| 243 | + 0 |
| 244 | + )); |
| 245 | + if (m.accent == 1) { |
| 246 | + accentPosition = moraIndex; |
| 247 | + } |
| 248 | + } |
| 249 | + } |
| 250 | + if (_moras.isEmpty()) { |
| 251 | + continue; |
| 252 | + } |
| 253 | + result.add(new AccentPhrase( |
| 254 | + _moras.toArray(Mora[]::new), |
| 255 | + accentPosition + 1, |
| 256 | + false, |
| 257 | + null |
| 258 | + )); |
| 259 | + } |
| 260 | + return result.toArray(AccentPhrase[]::new); |
| 261 | + }; |
| 262 | + |
| 263 | + Function<AccentPhrase[], Detail[]> accentPhrasesToProsody = accentPhrases -> { |
| 264 | + return Arrays.stream(accentPhrases).map(accentPhrase -> { |
| 265 | + List<Detail> detail = new ArrayList<>(); |
| 266 | + |
| 267 | + int i = 0; |
| 268 | + Arrays.stream(accentPhrase.moras).forEach(mora -> { |
| 269 | + String phoneme; |
| 270 | + if (mora.consonant != null && !mora.consonant.isEmpty()) { |
| 271 | + phoneme = mora.consonant + "-" + mora.vowel; |
| 272 | + } else { |
| 273 | + phoneme = mora.vowel; |
| 274 | + } |
| 275 | + |
| 276 | + var accent = 0; |
| 277 | + if (i == accentPhrase.accent - 1 || (i != 0 && i <= accentPhrase.accent - 1)) { |
| 278 | + accent = 1; |
| 279 | + } |
| 280 | + |
| 281 | + detail.add(new Detail( |
| 282 | + CharNormalizerJa.ToHiragana.normalize(mora.text), |
| 283 | + phoneme, |
| 284 | + accent |
| 285 | + )); |
| 286 | + }); |
| 287 | + |
| 288 | + if (accentPhrase.pause_mora != null) { |
| 289 | + detail.add(new Detail( |
| 290 | + "、", |
| 291 | + "_", |
| 292 | + 0 |
| 293 | + )); |
| 294 | + } |
| 295 | + |
| 296 | + return detail; |
| 297 | + }).flatMap(Collection::stream).toArray(Detail[]::new); |
| 298 | + }; |
| 299 | + |
| 300 | + public static class Synthesis { |
| 301 | + public String speakerUuid; |
| 302 | + public int styleId; |
| 303 | + public String text; |
| 304 | + public Detail[][] prosodyDetail; |
| 305 | + /** 0.5 ~ 1 ~ 2 */ |
| 306 | + public float speedScale = 1; |
| 307 | + /** 0 ~ 1 ~ +2 */ |
| 308 | + public float volumeScale = 1; |
| 309 | + /** -0.15 ~ 0 ~ +0.15 */ |
| 310 | + public float pitchScale = 0; |
| 311 | + /** 0 ~ 1 ~ +2 */ |
| 312 | + public float intonationScale = 1; |
| 313 | + /** 0 ~ 0.1 ~ +1.50 */ |
| 314 | + public float prePhonemeLength = 0.1f; |
| 315 | + /** 0 ~ 0.1 ~ +1.50 */ |
| 316 | + public float postPhonemeLength = 0.1f; |
| 317 | + public int outputSamplingRate = 22050; |
| 318 | + } |
| 319 | + |
| 320 | + public static Response execWithBackoff(Callable<Response> i) { |
| 321 | + ExponentialBackOff backoff = new ExponentialBackOff.Builder().build(); |
| 322 | + |
| 323 | + long delay = 0; |
| 324 | + |
| 325 | + Response response; |
| 326 | + do { |
| 327 | + try { |
| 328 | + Thread.sleep(delay); |
| 329 | + |
| 330 | + response = i.call(); |
| 331 | + |
| 332 | + if (response.getStatusInfo().getFamily() == Family.SERVER_ERROR) { |
| 333 | + logger.warning(String.format("Server error %s when accessing path %s. Delaying %dms", response.getStatus(), response.getLocation() != null ? response.getLocation().toASCIIString() : "null", delay)); |
| 334 | + } |
| 335 | + |
| 336 | + delay = backoff.nextBackOffMillis(); |
| 337 | + } catch (Exception e) { //callable throws exception |
| 338 | + throw new RuntimeException("Client request failed", e); |
| 339 | + } |
| 340 | + |
| 341 | + } while (delay != ExponentialBackOff.STOP && response.getStatusInfo().getFamily() == Family.SERVER_ERROR); |
| 342 | + |
| 343 | + if (response.getStatusInfo().getFamily() == Family.SERVER_ERROR) { |
| 344 | + throw new IllegalStateException("Client request failed for " + response.getLocation().toASCIIString()); |
| 345 | + } |
| 346 | + |
| 347 | + return response; |
| 348 | + } |
| 349 | +} |
0 commit comments