mirror of
https://git.undefined-dev.fr/Blendman974/faceclaw.git
synced 2026-11-19 21:03:09 +00:00
Optimize APK size
Removes unused keyword-spotting model. Makes on-device voice model a runtime download rather than bundling it. Removes native-code library versions for 32-bit and x86_64 processors.
This commit is contained in:
1 parent
69ec828cb7
commit
2b71eddab5
27 files changed
+313
-1283
No files matched your search
@@ -16,10 +16,8 @@ dependencies {
|
||||
|
||||
def faceclawSherpaOnnxVersion = "1.13.0"
|
||||
def faceclawLibLc3Version = "1.1.3"
|
||||
def faceclawAsrModelName = "sherpa-onnx-moonshine-base-en-quantized-2026-02-27"
|
||||
def faceclawAbi = "arm64-v8a"
|
||||
def faceclawJniLibsDir = file("$projectDir/src/main/jniLibs/$faceclawAbi")
|
||||
def faceclawAssetsDir = file("$projectDir/src/main/assets/faceclaw-voice-asr")
|
||||
def faceclawNativeBuildDir = file("$buildDir/faceclaw-native")
|
||||
|
||||
def faceclawDownload = { String url, File destination ->
|
||||
@@ -67,15 +65,11 @@ tasks.register("prepareFaceclawNativeLibs") {
|
||||
outputs.files(
|
||||
file("$faceclawJniLibsDir/libonnxruntime.so"),
|
||||
file("$faceclawJniLibsDir/libsherpa-onnx-jni.so"),
|
||||
file("$faceclawJniLibsDir/libfaceclaw_lc3.so"),
|
||||
file("$faceclawAssetsDir/$faceclawAsrModelName/encoder_model.ort"),
|
||||
file("$faceclawAssetsDir/$faceclawAsrModelName/decoder_model_merged.ort"),
|
||||
file("$faceclawAssetsDir/$faceclawAsrModelName/tokens.txt")
|
||||
file("$faceclawJniLibsDir/libfaceclaw_lc3.so")
|
||||
)
|
||||
|
||||
doLast {
|
||||
faceclawJniLibsDir.mkdirs()
|
||||
faceclawAssetsDir.mkdirs()
|
||||
|
||||
def downloadsDir = file("$faceclawNativeBuildDir/downloads")
|
||||
def sherpaArchive = file("$downloadsDir/sherpa-onnx-v${faceclawSherpaOnnxVersion}-android.tar.bz2")
|
||||
@@ -92,20 +86,6 @@ tasks.register("prepareFaceclawNativeLibs") {
|
||||
into faceclawJniLibsDir
|
||||
}
|
||||
|
||||
def asrArchive = file("$downloadsDir/${faceclawAsrModelName}.tar.bz2")
|
||||
faceclawDownload(
|
||||
"https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/${faceclawAsrModelName}.tar.bz2",
|
||||
asrArchive
|
||||
)
|
||||
copy {
|
||||
from tarTree(resources.bzip2(asrArchive))
|
||||
include "$faceclawAsrModelName/encoder_model.ort"
|
||||
include "$faceclawAsrModelName/decoder_model_merged.ort"
|
||||
include "$faceclawAsrModelName/tokens.txt"
|
||||
includeEmptyDirs = false
|
||||
into faceclawAssetsDir
|
||||
}
|
||||
|
||||
def lc3Archive = file("$downloadsDir/liblc3-v${faceclawLibLc3Version}.tar.gz")
|
||||
faceclawDownload(
|
||||
"https://github.com/google/liblc3/archive/refs/tags/v${faceclawLibLc3Version}.tar.gz",
|
||||
@@ -281,6 +261,11 @@ android {
|
||||
minSdkVersion 24
|
||||
targetSdkVersion 35
|
||||
ndk {
|
||||
// The NativeScript platform build.gradle has already added all four
|
||||
// ABIs by the time app.gradle is applied, and abiFilters is additive,
|
||||
// so the set must be cleared for this filter to take effect. Only
|
||||
// arm64 devices are supported (all our own JNI libs are arm64-only).
|
||||
abiFilters.clear()
|
||||
abiFilters "arm64-v8a"
|
||||
}
|
||||
|
||||
|
||||
-16
@@ -1,16 +0,0 @@
|
||||
---
|
||||
frameworks:
|
||||
- 其他
|
||||
license: Apache License 2.0
|
||||
tasks:
|
||||
- keyword-spotting
|
||||
---
|
||||
|
||||
sherpa 中自定义唤醒词模型,训练数据为 gigaspeech XL (10000 小时)数据,模型大小约为 3.3 M, 建模单元为 BPE 。 使用 icefall 训练,已转换为 onnx 格式,此仓库主要是给 sherpa-onnx 这个 inference 引擎使用。
|
||||
|
||||
模型结构为 zipformer 模型, 本质是一个非常小的语音识别模型,为了实现唤醒词的功能,我们在解码端做了一些修改和约束。 支持自定义唤醒词,数量不限,效果需要单个调整参数。
|
||||
|
||||
#### Clone with HTTP
|
||||
```bash
|
||||
git clone https://www.modelscope.cn/pkufool/sherpa-onnx-kws-zipformer-gigaspeech-3.3M-2024-01-01.git
|
||||
```
|
||||
BIN
Binary file not shown.
BIN
Binary file not shown.
BIN
Binary file not shown.
BIN
Binary file not shown.
-10
@@ -1,10 +0,0 @@
|
||||
▁HE LL O ▁WORLD
|
||||
▁HI ▁GO O G LE
|
||||
▁HE Y ▁S I RI
|
||||
▁A LE X A
|
||||
▁LOVE ▁AND ▁P E A CE
|
||||
▁PLAY ▁MU S IC
|
||||
▁GO ▁HOME
|
||||
▁HA PP Y ▁NEW ▁YEAR
|
||||
▁ME R RY ▁ CH R IST MA S
|
||||
▁S C RE EN ▁ON :1.5 #0.35
|
||||
-9
@@ -1,9 +0,0 @@
|
||||
HELLO WORLD
|
||||
HI GOOGLE
|
||||
HEY SIRI
|
||||
ALEXA
|
||||
LOVE AND PEACE
|
||||
PLAY MUSIC
|
||||
GO HOME
|
||||
HAPPY NEW YEAR
|
||||
MERRY CHRISTMAS
|
||||
-1
@@ -1 +0,0 @@
|
||||
▁S C RE EN ▁ON :1.5 #0.35
|
||||
-500
@@ -1,500 +0,0 @@
|
||||
<blk> 0
|
||||
<sos/eos> 1
|
||||
<unk> 2
|
||||
S 3
|
||||
T 4
|
||||
▁THE 5
|
||||
▁A 6
|
||||
▁TO 7
|
||||
▁AND 8
|
||||
N 9
|
||||
E 10
|
||||
D 11
|
||||
ING 12
|
||||
' 13
|
||||
▁OF 14
|
||||
▁I 15
|
||||
▁IN 16
|
||||
Y 17
|
||||
ED 18
|
||||
M 19
|
||||
▁ 20
|
||||
▁THAT 21
|
||||
O 22
|
||||
C 23
|
||||
R 24
|
||||
A 25
|
||||
P 26
|
||||
▁YOU 27
|
||||
U 28
|
||||
RE 29
|
||||
▁IT 30
|
||||
▁S 31
|
||||
G 32
|
||||
L 33
|
||||
B 34
|
||||
▁IS 35
|
||||
I 36
|
||||
AL 37
|
||||
ER 38
|
||||
▁RE 39
|
||||
▁C 40
|
||||
▁WE 41
|
||||
▁SO 42
|
||||
▁BE 43
|
||||
OR 44
|
||||
AR 45
|
||||
IN 46
|
||||
▁FOR 47
|
||||
F 48
|
||||
▁HE 49
|
||||
LE 50
|
||||
LY 51
|
||||
▁E 52
|
||||
▁WAS 53
|
||||
K 54
|
||||
▁F 55
|
||||
W 56
|
||||
VE 57
|
||||
RI 58
|
||||
CH 59
|
||||
LL 60
|
||||
▁THIS 61
|
||||
▁T 62
|
||||
EN 63
|
||||
▁ON 64
|
||||
IT 65
|
||||
CE 66
|
||||
▁WITH 67
|
||||
▁B 68
|
||||
▁W 69
|
||||
▁ME 70
|
||||
▁DO 71
|
||||
▁DE 72
|
||||
▁HAVE 73
|
||||
▁BUT 74
|
||||
ENT 75
|
||||
▁P 76
|
||||
UR 77
|
||||
ON 78
|
||||
▁AS 79
|
||||
IC 80
|
||||
TER 81
|
||||
TE 82
|
||||
▁THEY 83
|
||||
ES 84
|
||||
RA 85
|
||||
▁NOT 86
|
||||
TH 87
|
||||
▁ARE 88
|
||||
RO 89
|
||||
IL 90
|
||||
ATION 91
|
||||
AN 92
|
||||
▁ST 93
|
||||
▁WHAT 94
|
||||
SE 95
|
||||
LI 96
|
||||
V 97
|
||||
▁CAN 98
|
||||
▁G 99
|
||||
▁OR 100
|
||||
▁ALL 101
|
||||
NE 102
|
||||
▁AT 103
|
||||
▁CON 104
|
||||
H 105
|
||||
▁LIKE 106
|
||||
▁MO 107
|
||||
▁HIS 108
|
||||
▁AN 109
|
||||
▁MA 110
|
||||
▁THERE 111
|
||||
▁SU 112
|
||||
▁MY 113
|
||||
NG 114
|
||||
- 115
|
||||
▁ABOUT 116
|
||||
▁SE 117
|
||||
LA 118
|
||||
▁ONE 119
|
||||
LO 120
|
||||
▁SHE 121
|
||||
CK 122
|
||||
▁HAD 123
|
||||
AD 124
|
||||
MENT 125
|
||||
VER 126
|
||||
▁CA 127
|
||||
IR 128
|
||||
▁IF 129
|
||||
▁NO 130
|
||||
EL 131
|
||||
▁O 132
|
||||
▁FROM 133
|
||||
▁PA 134
|
||||
▁YOUR 135
|
||||
▁KNOW 136
|
||||
RY 137
|
||||
ION 138
|
||||
UN 139
|
||||
▁JUST 140
|
||||
▁UP 141
|
||||
▁US 142
|
||||
▁WHO 143
|
||||
▁GO 144
|
||||
ME 145
|
||||
▁EX 146
|
||||
▁OUT 147
|
||||
ND 148
|
||||
CI 149
|
||||
▁PO 150
|
||||
▁HER 151
|
||||
▁FA 152
|
||||
▁PRO 153
|
||||
▁WHEN 154
|
||||
ATE 155
|
||||
AT 156
|
||||
▁BY 157
|
||||
PE 158
|
||||
AS 159
|
||||
▁LA 160
|
||||
GE 161
|
||||
▁CO 162
|
||||
▁PEOPLE 163
|
||||
▁K 164
|
||||
▁SOME 165
|
||||
IGHT 166
|
||||
▁THINK 167
|
||||
VI 168
|
||||
DE 169
|
||||
▁LE 170
|
||||
▁WERE 171
|
||||
▁LI 172
|
||||
OM 173
|
||||
▁WOULD 174
|
||||
▁SEE 175
|
||||
CO 176
|
||||
▁BO 177
|
||||
▁OUR 178
|
||||
▁HOW 179
|
||||
ET 180
|
||||
Z 181
|
||||
UL 182
|
||||
▁HO 183
|
||||
IG 184
|
||||
▁GET 185
|
||||
▁DON 186
|
||||
IS 187
|
||||
URE 188
|
||||
▁WHICH 189
|
||||
▁MI 190
|
||||
IES 191
|
||||
US 192
|
||||
X 193
|
||||
▁MORE 194
|
||||
▁NOW 195
|
||||
▁TIME 196
|
||||
▁UN 197
|
||||
▁FI 198
|
||||
TY 199
|
||||
▁WILL 200
|
||||
RU 201
|
||||
IVE 202
|
||||
▁THEIR 203
|
||||
▁SAY 204
|
||||
ITY 205
|
||||
▁HA 206
|
||||
LU 207
|
||||
▁RO 208
|
||||
HO 209
|
||||
OL 210
|
||||
IST 211
|
||||
AGE 212
|
||||
MA 213
|
||||
KE 214
|
||||
▁RA 215
|
||||
ID 216
|
||||
ERS 217
|
||||
▁LO 218
|
||||
▁SP 219
|
||||
▁DA 220
|
||||
▁SAID 221
|
||||
TION 222
|
||||
THER 223
|
||||
▁EN 224
|
||||
▁BECAUSE 225
|
||||
▁HAS 226
|
||||
EST 227
|
||||
TI 228
|
||||
ACT 229
|
||||
▁BA 230
|
||||
▁VERY 231
|
||||
▁THEM 232
|
||||
▁BEEN 233
|
||||
▁WA 234
|
||||
ABLE 235
|
||||
▁SA 236
|
||||
▁DID 237
|
||||
▁HIM 238
|
||||
▁DIS 239
|
||||
▁WANT 240
|
||||
▁OTHER 241
|
||||
IM 242
|
||||
▁PRE 243
|
||||
▁LOOK 244
|
||||
▁GOING 245
|
||||
▁VI 246
|
||||
▁WORK 247
|
||||
▁DR 248
|
||||
ENCE 249
|
||||
▁DI 250
|
||||
FF 251
|
||||
TA 252
|
||||
▁REALLY 253
|
||||
▁RIGHT 254
|
||||
MO 255
|
||||
ALLY 256
|
||||
▁INTO 257
|
||||
OW 258
|
||||
▁WAY 259
|
||||
▁FE 260
|
||||
▁COULD 261
|
||||
▁THEN 262
|
||||
TED 263
|
||||
▁HERE 264
|
||||
▁THESE 265
|
||||
▁OVER 266
|
||||
▁WELL 267
|
||||
▁NA 268
|
||||
▁ANY 269
|
||||
HE 270
|
||||
▁COMP 271
|
||||
▁COM 272
|
||||
▁TH 273
|
||||
▁EVEN 274
|
||||
▁TWO 275
|
||||
▁JO 276
|
||||
▁MAKE 277
|
||||
▁BACK 278
|
||||
▁WHERE 279
|
||||
SH 280
|
||||
▁QU 281
|
||||
MAN 282
|
||||
▁YEAR 283
|
||||
▁TRA 284
|
||||
▁CHA 285
|
||||
QU 286
|
||||
CA 287
|
||||
▁ALSO 288
|
||||
CTION 289
|
||||
ANCE 290
|
||||
UM 291
|
||||
PER 292
|
||||
DI 293
|
||||
▁HU 294
|
||||
▁MU 295
|
||||
FUL 296
|
||||
▁NEW 297
|
||||
ARD 298
|
||||
J 299
|
||||
ICAL 300
|
||||
ANT 301
|
||||
END 302
|
||||
▁MAR 303
|
||||
▁NE 304
|
||||
IF 305
|
||||
DER 306
|
||||
ONE 307
|
||||
▁THOSE 308
|
||||
PP 309
|
||||
▁MAN 310
|
||||
▁GOOD 311
|
||||
▁TAKE 312
|
||||
SIDE 313
|
||||
PO 314
|
||||
HA 315
|
||||
▁LITTLE 316
|
||||
▁APP 317
|
||||
▁BI 318
|
||||
UGH 319
|
||||
▁COME 320
|
||||
ISH 321
|
||||
TIC 322
|
||||
▁PER 323
|
||||
▁THINGS 324
|
||||
▁MUCH 325
|
||||
▁FIRST 326
|
||||
▁THAN 327
|
||||
IA 328
|
||||
▁AGAIN 329
|
||||
▁STA 330
|
||||
▁PART 331
|
||||
▁GA 332
|
||||
▁MEAN 333
|
||||
▁LET 334
|
||||
▁DOWN 335
|
||||
▁NEED 336
|
||||
▁GU 337
|
||||
LES 338
|
||||
▁RU 339
|
||||
ATED 340
|
||||
▁SHA 341
|
||||
VO 342
|
||||
▁TALK 343
|
||||
▁DAY 344
|
||||
▁CHI 345
|
||||
▁SOMETHING 346
|
||||
▁AFTER 347
|
||||
▁THOUGH 348
|
||||
▁CALL 349
|
||||
▁GRA 350
|
||||
▁EVERY 351
|
||||
▁KIND 352
|
||||
ROW 353
|
||||
AIN 354
|
||||
▁BU 355
|
||||
▁THROUGH 356
|
||||
▁SHOULD 357
|
||||
▁ONLY 358
|
||||
▁TA 359
|
||||
▁MIN 360
|
||||
▁GOT 361
|
||||
IAN 362
|
||||
▁GREAT 363
|
||||
▁RI 364
|
||||
▁DOES 365
|
||||
▁THING 366
|
||||
DAY 367
|
||||
OUS 368
|
||||
▁START 369
|
||||
▁MOST 370
|
||||
▁COL 371
|
||||
PH 372
|
||||
▁LOT 373
|
||||
IZE 374
|
||||
BER 375
|
||||
IOUS 376
|
||||
▁LIFE 377
|
||||
▁MANY 378
|
||||
▁FEEL 379
|
||||
ITION 380
|
||||
▁VA 381
|
||||
NESS 382
|
||||
▁BEFORE 383
|
||||
▁THREE 384
|
||||
▁WORLD 385
|
||||
▁GIVE 386
|
||||
▁WHY 387
|
||||
▁PI 388
|
||||
▁LONG 389
|
||||
PORT 390
|
||||
▁SHOW 391
|
||||
▁ACTUALLY 392
|
||||
▁NEVER 393
|
||||
IV 394
|
||||
FORM 395
|
||||
▁THANK 396
|
||||
▁HI 397
|
||||
▁PUT 398
|
||||
▁LOVE 399
|
||||
▁QUESTION 400
|
||||
▁HELP 401
|
||||
▁HAPPEN 402
|
||||
▁MR 403
|
||||
▁OWN 404
|
||||
WARD 405
|
||||
▁MADE 406
|
||||
▁HAND 407
|
||||
▁STATE 408
|
||||
IBLE 409
|
||||
▁TELL 410
|
||||
▁PH 411
|
||||
▁CAME 412
|
||||
▁TEA 413
|
||||
▁STILL 414
|
||||
▁DIFFERENT 415
|
||||
▁PLACE 416
|
||||
▁MIGHT 417
|
||||
▁FIND 418
|
||||
▁DU 419
|
||||
▁IMP 420
|
||||
▁OLD 421
|
||||
▁PERSON 422
|
||||
▁COUNT 423
|
||||
▁JA 424
|
||||
▁POINT 425
|
||||
▁HIGH 426
|
||||
SHIP 427
|
||||
▁LAST 428
|
||||
▁SAME 429
|
||||
▁FOUR 430
|
||||
▁AROUND 431
|
||||
▁OH 432
|
||||
▁CHANGE 433
|
||||
▁UNDER 434
|
||||
▁HOME 435
|
||||
▁ANOTHER 436
|
||||
▁FOUND 437
|
||||
ATIVE 438
|
||||
▁TURN 439
|
||||
▁YES 440
|
||||
LAND 441
|
||||
▁ALWAYS 442
|
||||
▁YEAH 443
|
||||
▁SCHOOL 444
|
||||
▁JE 445
|
||||
▁NEXT 446
|
||||
▁WORD 447
|
||||
▁JU 448
|
||||
LINE 449
|
||||
▁SIX 450
|
||||
▁LEARN 451
|
||||
▁BOOK 452
|
||||
▁PLAY 453
|
||||
▁AWAY 454
|
||||
▁HOUSE 455
|
||||
▁PLAN 456
|
||||
▁WHILE 457
|
||||
▁COMMUNI 458
|
||||
▁FRIEND 459
|
||||
▁AMERICA 460
|
||||
▁INTEREST 461
|
||||
▁KEEP 462
|
||||
▁BETTER 463
|
||||
▁IDEA 464
|
||||
▁HARD 465
|
||||
▁TWENTY 466
|
||||
▁THOUSAND 467
|
||||
▁HUNDRED 468
|
||||
▁WEEK 469
|
||||
▁NUMBER 470
|
||||
▁COURSE 471
|
||||
▁OKAY 472
|
||||
PRESS 473
|
||||
▁SYSTEM 474
|
||||
▁NINE 475
|
||||
▁UNDERSTAND 476
|
||||
▁WITHOUT 477
|
||||
▁BELIEVE 478
|
||||
▁POWER 479
|
||||
▁MAYBE 480
|
||||
▁BUILD 481
|
||||
▁MONEY 482
|
||||
▁BETWEEN 483
|
||||
▁GONNA 484
|
||||
▁IMPORTANT 485
|
||||
▁ANYTHING 486
|
||||
▁LEFT 487
|
||||
▁HAVING 488
|
||||
▁WHITE 489
|
||||
▁PROBLEM 490
|
||||
▁SECOND 491
|
||||
▁SPEAK 492
|
||||
▁TOGETHER 493
|
||||
▁BLACK 494
|
||||
▁PUBLIC 495
|
||||
▁HEALTH 496
|
||||
▁NOTHING 497
|
||||
▁EVERYTHING 498
|
||||
Q 499
|
||||
+52
-147
@@ -1,30 +1,20 @@
|
||||
package com.faceclaw.app;
|
||||
|
||||
import android.content.Context;
|
||||
import android.content.res.AssetManager;
|
||||
import android.os.Handler;
|
||||
import android.os.Looper;
|
||||
import android.os.SystemClock;
|
||||
import android.util.Log;
|
||||
|
||||
import com.k2fsa.sherpa.onnx.FeatureConfig;
|
||||
import com.k2fsa.sherpa.onnx.KeywordSpotter;
|
||||
import com.k2fsa.sherpa.onnx.KeywordSpotterConfig;
|
||||
import com.k2fsa.sherpa.onnx.KeywordSpotterResult;
|
||||
import com.k2fsa.sherpa.onnx.OfflineModelConfig;
|
||||
import com.k2fsa.sherpa.onnx.OfflineMoonshineModelConfig;
|
||||
import com.k2fsa.sherpa.onnx.OfflineRecognizer;
|
||||
import com.k2fsa.sherpa.onnx.OfflineRecognizerConfig;
|
||||
import com.k2fsa.sherpa.onnx.OfflineRecognizerResult;
|
||||
import com.k2fsa.sherpa.onnx.OfflineStream;
|
||||
import com.k2fsa.sherpa.onnx.OnlineModelConfig;
|
||||
import com.k2fsa.sherpa.onnx.OnlineStream;
|
||||
import com.k2fsa.sherpa.onnx.OnlineTransducerModelConfig;
|
||||
|
||||
import java.io.File;
|
||||
import java.io.FileOutputStream;
|
||||
import java.io.IOException;
|
||||
import java.io.InputStream;
|
||||
import java.nio.charset.StandardCharsets;
|
||||
import java.util.ArrayDeque;
|
||||
import java.util.Arrays;
|
||||
@@ -57,17 +47,15 @@ public class FaceclawVoiceController {
|
||||
private static final float TRANSCRIPT_NORMALIZE_TARGET_PEAK = 0.9f;
|
||||
private static final float TRANSCRIPT_NORMALIZE_MAX_GAIN = 30f;
|
||||
private static final int TRANSCRIPT_LOG_PREVIEW_CHARS = 80;
|
||||
private static final String ASSET_ROOT = "faceclaw-voice";
|
||||
private static final String MODEL_DIR = "sherpa-onnx-kws-zipformer-gigaspeech-3.3M-2024-01-01";
|
||||
private static final String ASR_ASSET_ROOT = "faceclaw-voice-asr";
|
||||
// Model directory shared with the TS-side download flow (asr-model.ts),
|
||||
// which fetches the Moonshine files here on demand; they are no longer
|
||||
// bundled in the APK.
|
||||
private static final String ASR_ROOT = "faceclaw-voice-asr";
|
||||
private static final String ASR_MODEL_DIR = "sherpa-onnx-moonshine-base-en-quantized-2026-02-27";
|
||||
private static final String[] MODEL_FILES = {
|
||||
"encoder-epoch-12-avg-2-chunk-16-left-64.onnx",
|
||||
"decoder-epoch-12-avg-2-chunk-16-left-64.onnx",
|
||||
"joiner-epoch-12-avg-2-chunk-16-left-64.onnx",
|
||||
"tokens.txt",
|
||||
"screen-on-keywords.txt"
|
||||
};
|
||||
// Model files for the retired on-phone wake-word spotter, copied to
|
||||
// filesDir by earlier releases; deleted on sight to reclaim the space.
|
||||
// (The wakeword is now detected by the glasses firmware itself.)
|
||||
private static final String LEGACY_KWS_ROOT = "faceclaw-voice";
|
||||
private static final String[] ASR_MODEL_FILES = {
|
||||
"encoder_model.ort",
|
||||
"decoder_model_merged.ort",
|
||||
@@ -75,7 +63,6 @@ public class FaceclawVoiceController {
|
||||
};
|
||||
|
||||
private enum VoiceInputMode {
|
||||
WAKEWORD, // on-phone keyword spotting (kept for later; not currently wired)
|
||||
ONBOARD, // on-phone Moonshine transcription
|
||||
CLOUD // decode locally, emit PCM for a cloud recognizer on the TS side
|
||||
}
|
||||
@@ -89,10 +76,8 @@ public class FaceclawVoiceController {
|
||||
private volatile FaceclawBleCommunicator communicator;
|
||||
private Thread workerThread;
|
||||
private volatile boolean started;
|
||||
private VoiceInputMode mode = VoiceInputMode.WAKEWORD;
|
||||
private KeywordSpotter keywordSpotter;
|
||||
private VoiceInputMode mode = VoiceInputMode.CLOUD;
|
||||
private OfflineRecognizer recognizer;
|
||||
private OnlineStream stream;
|
||||
private FaceclawLc3Decoder lc3Decoder;
|
||||
private final float[] transcriptSamples = new float[TRANSCRIPT_SEGMENT_MAX_SAMPLES];
|
||||
private int transcriptSampleCount;
|
||||
@@ -140,10 +125,6 @@ public class FaceclawVoiceController {
|
||||
this.endpointing = endpointing;
|
||||
}
|
||||
|
||||
public void start() {
|
||||
start("wakeword");
|
||||
}
|
||||
|
||||
public void start(String requestedMode) {
|
||||
synchronized (lock) {
|
||||
if (started) {
|
||||
@@ -194,26 +175,23 @@ public class FaceclawVoiceController {
|
||||
if ("cloud".equals(requestedMode)) {
|
||||
return VoiceInputMode.CLOUD;
|
||||
}
|
||||
if ("onboard".equals(requestedMode) || "full".equals(requestedMode)) {
|
||||
return VoiceInputMode.ONBOARD;
|
||||
}
|
||||
return VoiceInputMode.WAKEWORD;
|
||||
return VoiceInputMode.ONBOARD;
|
||||
}
|
||||
|
||||
private void runLoop() {
|
||||
try {
|
||||
deleteLegacyKwsFiles();
|
||||
VoiceInputMode currentMode = mode;
|
||||
if (currentMode == VoiceInputMode.ONBOARD) {
|
||||
File modelDir = findAsrModelDir();
|
||||
if (modelDir == null) {
|
||||
emitStatus("Voice model not downloaded (see Settings > Voice).");
|
||||
return;
|
||||
}
|
||||
emitStatus("Loading transcription model...");
|
||||
File modelDir = installAsrModelFiles();
|
||||
recognizer = new OfflineRecognizer(buildRecognizerConfig(modelDir));
|
||||
resetTranscriptState();
|
||||
lastTranscript = "";
|
||||
} else if (currentMode == VoiceInputMode.WAKEWORD) {
|
||||
emitStatus("Loading wake-word model...");
|
||||
File modelDir = installModelFiles();
|
||||
keywordSpotter = new KeywordSpotter(buildConfig(modelDir));
|
||||
stream = keywordSpotter.createStream();
|
||||
}
|
||||
lc3Decoder = new FaceclawLc3Decoder();
|
||||
endpointDetector.reset();
|
||||
@@ -225,9 +203,7 @@ public class FaceclawVoiceController {
|
||||
|
||||
emitStatus(currentMode == VoiceInputMode.CLOUD
|
||||
? "Listening (cloud)..."
|
||||
: currentMode == VoiceInputMode.ONBOARD
|
||||
? "Listening..."
|
||||
: "Listening for \"screen on\"...");
|
||||
: "Listening...");
|
||||
processG2Audio();
|
||||
// Button released / stop requested: emit one final full-utterance
|
||||
// transcript so the UI can freeze it.
|
||||
@@ -312,29 +288,6 @@ public class FaceclawVoiceController {
|
||||
return b.array();
|
||||
}
|
||||
|
||||
private KeywordSpotterConfig buildConfig(File modelDir) {
|
||||
return KeywordSpotterConfig.builder()
|
||||
.setFeatureConfig(FeatureConfig.builder()
|
||||
.setSampleRate(SAMPLE_RATE)
|
||||
.setFeatureDim(FEATURE_DIM)
|
||||
.build())
|
||||
.setOnlineModelConfig(OnlineModelConfig.builder()
|
||||
.setTransducer(OnlineTransducerModelConfig.builder()
|
||||
.setEncoder(new File(modelDir, "encoder-epoch-12-avg-2-chunk-16-left-64.onnx").getAbsolutePath())
|
||||
.setDecoder(new File(modelDir, "decoder-epoch-12-avg-2-chunk-16-left-64.onnx").getAbsolutePath())
|
||||
.setJoiner(new File(modelDir, "joiner-epoch-12-avg-2-chunk-16-left-64.onnx").getAbsolutePath())
|
||||
.build())
|
||||
.setTokens(new File(modelDir, "tokens.txt").getAbsolutePath())
|
||||
.setModelType("zipformer2")
|
||||
.setModelingUnit("")
|
||||
.setNumThreads(1)
|
||||
.build())
|
||||
.setKeywordsFile(new File(modelDir, "screen-on-keywords.txt").getAbsolutePath())
|
||||
.setKeywordsScore(1.5f)
|
||||
.setKeywordsThreshold(0.35f)
|
||||
.build();
|
||||
}
|
||||
|
||||
private OfflineRecognizerConfig buildRecognizerConfig(File modelDir) {
|
||||
return OfflineRecognizerConfig.builder()
|
||||
.setFeatureConfig(FeatureConfig.builder()
|
||||
@@ -352,50 +305,42 @@ public class FaceclawVoiceController {
|
||||
.build();
|
||||
}
|
||||
|
||||
private File installModelFiles() throws IOException {
|
||||
File modelDir = new File(appContext.getFilesDir(), ASSET_ROOT + File.separator + MODEL_DIR);
|
||||
if (!modelDir.exists() && !modelDir.mkdirs()) {
|
||||
throw new IOException("Could not create " + modelDir.getAbsolutePath());
|
||||
}
|
||||
AssetManager assets = appContext.getAssets();
|
||||
for (String fileName : MODEL_FILES) {
|
||||
copyAssetIfNeeded(
|
||||
assets,
|
||||
ASSET_ROOT + "/" + MODEL_DIR + "/" + fileName,
|
||||
new File(modelDir, fileName)
|
||||
);
|
||||
}
|
||||
return modelDir;
|
||||
}
|
||||
|
||||
private File installAsrModelFiles() throws IOException {
|
||||
File modelDir = new File(appContext.getFilesDir(), ASR_ASSET_ROOT + File.separator + ASR_MODEL_DIR);
|
||||
if (!modelDir.exists() && !modelDir.mkdirs()) {
|
||||
throw new IOException("Could not create " + modelDir.getAbsolutePath());
|
||||
}
|
||||
AssetManager assets = appContext.getAssets();
|
||||
/**
|
||||
* The Moonshine model directory, populated by the download flow in
|
||||
* asr-model.ts (releases before 0.5.0 copied the same files out of the
|
||||
* APK, so upgraded installs are already complete). Null when any file is
|
||||
* missing, i.e. the model still needs to be downloaded.
|
||||
*/
|
||||
private File findAsrModelDir() {
|
||||
File modelDir = new File(appContext.getFilesDir(), ASR_ROOT + File.separator + ASR_MODEL_DIR);
|
||||
for (String fileName : ASR_MODEL_FILES) {
|
||||
copyAssetIfNeeded(
|
||||
assets,
|
||||
ASR_ASSET_ROOT + "/" + ASR_MODEL_DIR + "/" + fileName,
|
||||
new File(modelDir, fileName)
|
||||
);
|
||||
}
|
||||
return modelDir;
|
||||
}
|
||||
|
||||
private void copyAssetIfNeeded(AssetManager assets, String assetPath, File destination) throws IOException {
|
||||
if (destination.exists() && destination.length() > 0) {
|
||||
return;
|
||||
}
|
||||
try (InputStream input = assets.open(assetPath);
|
||||
FileOutputStream output = new FileOutputStream(destination)) {
|
||||
byte[] buffer = new byte[64 * 1024];
|
||||
int read;
|
||||
while ((read = input.read(buffer)) >= 0) {
|
||||
output.write(buffer, 0, read);
|
||||
File file = new File(modelDir, fileName);
|
||||
if (!file.exists() || file.length() == 0) {
|
||||
return null;
|
||||
}
|
||||
}
|
||||
return modelDir;
|
||||
}
|
||||
|
||||
private void deleteLegacyKwsFiles() {
|
||||
try {
|
||||
deleteRecursively(new File(appContext.getFilesDir(), LEGACY_KWS_ROOT));
|
||||
} catch (Throwable t) {
|
||||
Log.w(TAG, "failed to delete legacy wake-word files", t);
|
||||
}
|
||||
}
|
||||
|
||||
private static void deleteRecursively(File file) {
|
||||
if (!file.exists()) {
|
||||
return;
|
||||
}
|
||||
File[] children = file.listFiles();
|
||||
if (children != null) {
|
||||
for (File child : children) {
|
||||
deleteRecursively(child);
|
||||
}
|
||||
}
|
||||
file.delete();
|
||||
}
|
||||
|
||||
private boolean startG2Audio() {
|
||||
@@ -413,9 +358,8 @@ public class FaceclawVoiceController {
|
||||
private void processG2Audio() {
|
||||
short[] pcm = new short[FaceclawLc3Decoder.SAMPLES_PER_PACKET];
|
||||
while (started && !Thread.currentThread().isInterrupted()) {
|
||||
OnlineStream currentStream = stream;
|
||||
FaceclawLc3Decoder currentDecoder = lc3Decoder;
|
||||
if (currentDecoder == null || (mode == VoiceInputMode.WAKEWORD && currentStream == null)) {
|
||||
if (currentDecoder == null) {
|
||||
return;
|
||||
}
|
||||
|
||||
@@ -438,40 +382,17 @@ public class FaceclawVoiceController {
|
||||
}
|
||||
if (mode == VoiceInputMode.CLOUD) {
|
||||
emitPcm(pcm, count);
|
||||
} else if (mode == VoiceInputMode.ONBOARD) {
|
||||
float[] samples = new float[count];
|
||||
for (int i = 0; i < count; i++) {
|
||||
samples[i] = pcm[i] / 32768.0f;
|
||||
}
|
||||
processRecognizer(samples);
|
||||
} else {
|
||||
float[] samples = new float[count];
|
||||
for (int i = 0; i < count; i++) {
|
||||
samples[i] = pcm[i] / 32768.0f;
|
||||
}
|
||||
currentStream.acceptWaveform(samples, SAMPLE_RATE);
|
||||
processKeywordSpotter(currentStream);
|
||||
processRecognizer(samples);
|
||||
}
|
||||
maybeEmitAudioStats(false);
|
||||
}
|
||||
}
|
||||
|
||||
private void processKeywordSpotter(OnlineStream currentStream) {
|
||||
KeywordSpotter currentSpotter = keywordSpotter;
|
||||
if (currentSpotter == null) {
|
||||
return;
|
||||
}
|
||||
while (currentSpotter.isReady(currentStream)) {
|
||||
currentSpotter.decode(currentStream);
|
||||
KeywordSpotterResult result = currentSpotter.getResult(currentStream);
|
||||
String keyword = result == null ? "" : result.getKeyword();
|
||||
if (keyword != null && keyword.trim().length() > 0) {
|
||||
currentSpotter.reset(currentStream);
|
||||
emitWakeWord(keyword);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private void processRecognizer(float[] samples) {
|
||||
appendTranscriptSamples(samples);
|
||||
long now = SystemClock.elapsedRealtime();
|
||||
@@ -783,14 +704,6 @@ public class FaceclawVoiceController {
|
||||
}
|
||||
|
||||
private void releaseSherpa() {
|
||||
if (stream != null) {
|
||||
stream.release();
|
||||
stream = null;
|
||||
}
|
||||
if (keywordSpotter != null) {
|
||||
keywordSpotter.release();
|
||||
keywordSpotter = null;
|
||||
}
|
||||
if (recognizer != null) {
|
||||
recognizer.release();
|
||||
recognizer = null;
|
||||
@@ -893,14 +806,6 @@ public class FaceclawVoiceController {
|
||||
mainHandler.post(() -> currentListener.onStatus(status));
|
||||
}
|
||||
|
||||
private void emitWakeWord(String keyword) {
|
||||
FaceclawVoiceControllerListener currentListener = listener;
|
||||
if (currentListener == null) {
|
||||
return;
|
||||
}
|
||||
mainHandler.post(() -> currentListener.onWakeWord(keyword));
|
||||
}
|
||||
|
||||
private void emitTranscript(String text, boolean isFinal) {
|
||||
FaceclawVoiceControllerListener currentListener = listener;
|
||||
if (currentListener == null) {
|
||||
|
||||
-2
@@ -3,8 +3,6 @@ package com.faceclaw.app;
|
||||
public interface FaceclawVoiceControllerListener {
|
||||
void onStatus(String status);
|
||||
|
||||
void onWakeWord(String keyword);
|
||||
|
||||
/**
|
||||
* Best transcript of the current utterance so far. REPLACE semantics: text
|
||||
* is the complete transcript, not a delta — display it verbatim, replacing
|
||||
|
||||
@@ -1,66 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class KeywordSpotter {
|
||||
private long ptr;
|
||||
|
||||
public KeywordSpotter(KeywordSpotterConfig config) {
|
||||
System.loadLibrary("onnxruntime");
|
||||
System.loadLibrary("sherpa-onnx-jni");
|
||||
ptr = newFromFile(config);
|
||||
if (ptr == 0) {
|
||||
throw new IllegalArgumentException("Invalid KeywordSpotterConfig");
|
||||
}
|
||||
}
|
||||
|
||||
public OnlineStream createStream() {
|
||||
return createStream("");
|
||||
}
|
||||
|
||||
public OnlineStream createStream(String keywords) {
|
||||
return new OnlineStream(createStream(ptr, keywords == null ? "" : keywords));
|
||||
}
|
||||
|
||||
public boolean isReady(OnlineStream stream) {
|
||||
return stream != null && stream.getPtr() != 0 && isReady(ptr, stream.getPtr());
|
||||
}
|
||||
|
||||
public void decode(OnlineStream stream) {
|
||||
decode(ptr, stream.getPtr());
|
||||
}
|
||||
|
||||
public void reset(OnlineStream stream) {
|
||||
reset(ptr, stream.getPtr());
|
||||
}
|
||||
|
||||
public KeywordSpotterResult getResult(OnlineStream stream) {
|
||||
return getResult(ptr, stream.getPtr());
|
||||
}
|
||||
|
||||
public void release() {
|
||||
if (ptr == 0) {
|
||||
return;
|
||||
}
|
||||
delete(ptr);
|
||||
ptr = 0;
|
||||
}
|
||||
|
||||
@Override
|
||||
protected void finalize() throws Throwable {
|
||||
release();
|
||||
super.finalize();
|
||||
}
|
||||
|
||||
private native long newFromFile(KeywordSpotterConfig config);
|
||||
|
||||
private native void delete(long ptr);
|
||||
|
||||
private native long createStream(long ptr, String keywords);
|
||||
|
||||
private native boolean isReady(long ptr, long streamPtr);
|
||||
|
||||
private native void decode(long ptr, long streamPtr);
|
||||
|
||||
private native void reset(long ptr, long streamPtr);
|
||||
|
||||
private native KeywordSpotterResult getResult(long ptr, long streamPtr);
|
||||
}
|
||||
@@ -1,92 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class KeywordSpotterConfig {
|
||||
private final FeatureConfig featConfig;
|
||||
private final OnlineModelConfig modelConfig;
|
||||
private final int maxActivePaths;
|
||||
private final String keywordsFile;
|
||||
private final float keywordsScore;
|
||||
private final float keywordsThreshold;
|
||||
private final int numTrailingBlanks;
|
||||
|
||||
private KeywordSpotterConfig(Builder builder) {
|
||||
this.featConfig = builder.featConfig;
|
||||
this.modelConfig = builder.modelConfig;
|
||||
this.maxActivePaths = builder.maxActivePaths;
|
||||
this.keywordsFile = builder.keywordsFile;
|
||||
this.keywordsScore = builder.keywordsScore;
|
||||
this.keywordsThreshold = builder.keywordsThreshold;
|
||||
this.numTrailingBlanks = builder.numTrailingBlanks;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public FeatureConfig getFeatConfig() {
|
||||
return featConfig;
|
||||
}
|
||||
|
||||
public OnlineModelConfig getModelConfig() {
|
||||
return modelConfig;
|
||||
}
|
||||
|
||||
public int getMaxActivePaths() {
|
||||
return maxActivePaths;
|
||||
}
|
||||
|
||||
public String getKeywordsFile() {
|
||||
return keywordsFile;
|
||||
}
|
||||
|
||||
public float getKeywordsScore() {
|
||||
return keywordsScore;
|
||||
}
|
||||
|
||||
public float getKeywordsThreshold() {
|
||||
return keywordsThreshold;
|
||||
}
|
||||
|
||||
public int getNumTrailingBlanks() {
|
||||
return numTrailingBlanks;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private FeatureConfig featConfig = FeatureConfig.builder().build();
|
||||
private OnlineModelConfig modelConfig = OnlineModelConfig.builder().build();
|
||||
private int maxActivePaths = 4;
|
||||
private String keywordsFile = "keywords.txt";
|
||||
private float keywordsScore = 1.5f;
|
||||
private float keywordsThreshold = 0.25f;
|
||||
private int numTrailingBlanks = 2;
|
||||
|
||||
public Builder setFeatureConfig(FeatureConfig featConfig) {
|
||||
this.featConfig = featConfig;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setOnlineModelConfig(OnlineModelConfig modelConfig) {
|
||||
this.modelConfig = modelConfig;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setKeywordsFile(String keywordsFile) {
|
||||
this.keywordsFile = keywordsFile;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setKeywordsScore(float keywordsScore) {
|
||||
this.keywordsScore = keywordsScore;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setKeywordsThreshold(float keywordsThreshold) {
|
||||
this.keywordsThreshold = keywordsThreshold;
|
||||
return this;
|
||||
}
|
||||
|
||||
public KeywordSpotterConfig build() {
|
||||
return new KeywordSpotterConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -1,25 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class KeywordSpotterResult {
|
||||
private final String keyword;
|
||||
private final String[] tokens;
|
||||
private final float[] timestamps;
|
||||
|
||||
public KeywordSpotterResult(String keyword, String[] tokens, float[] timestamps) {
|
||||
this.keyword = keyword == null ? "" : keyword;
|
||||
this.tokens = tokens == null ? new String[0] : tokens;
|
||||
this.timestamps = timestamps == null ? new float[0] : timestamps;
|
||||
}
|
||||
|
||||
public String getKeyword() {
|
||||
return keyword;
|
||||
}
|
||||
|
||||
public String[] getTokens() {
|
||||
return tokens;
|
||||
}
|
||||
|
||||
public float[] getTimestamps() {
|
||||
return timestamps;
|
||||
}
|
||||
}
|
||||
@@ -1,137 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineModelConfig {
|
||||
private final OnlineTransducerModelConfig transducer;
|
||||
private final OnlineParaformerModelConfig paraformer;
|
||||
private final OnlineZipformer2CtcModelConfig zipformer2Ctc;
|
||||
private final OnlineNeMoCtcModelConfig neMoCtc;
|
||||
private final OnlineToneCtcModelConfig toneCtc;
|
||||
private final String tokens;
|
||||
private final int numThreads;
|
||||
private final boolean debug;
|
||||
private final String provider;
|
||||
private final String modelType;
|
||||
private final String modelingUnit;
|
||||
private final String bpeVocab;
|
||||
|
||||
private OnlineModelConfig(Builder builder) {
|
||||
this.transducer = builder.transducer;
|
||||
this.paraformer = builder.paraformer;
|
||||
this.zipformer2Ctc = builder.zipformer2Ctc;
|
||||
this.neMoCtc = builder.neMoCtc;
|
||||
this.toneCtc = builder.toneCtc;
|
||||
this.tokens = builder.tokens;
|
||||
this.numThreads = builder.numThreads;
|
||||
this.debug = builder.debug;
|
||||
this.provider = builder.provider;
|
||||
this.modelType = builder.modelType;
|
||||
this.modelingUnit = builder.modelingUnit;
|
||||
this.bpeVocab = builder.bpeVocab;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public OnlineTransducerModelConfig getTransducer() {
|
||||
return transducer;
|
||||
}
|
||||
|
||||
public OnlineParaformerModelConfig getParaformer() {
|
||||
return paraformer;
|
||||
}
|
||||
|
||||
public OnlineZipformer2CtcModelConfig getZipformer2Ctc() {
|
||||
return zipformer2Ctc;
|
||||
}
|
||||
|
||||
public OnlineNeMoCtcModelConfig getNeMoCtc() {
|
||||
return neMoCtc;
|
||||
}
|
||||
|
||||
public OnlineToneCtcModelConfig getToneCtc() {
|
||||
return toneCtc;
|
||||
}
|
||||
|
||||
public String getTokens() {
|
||||
return tokens;
|
||||
}
|
||||
|
||||
public int getNumThreads() {
|
||||
return numThreads;
|
||||
}
|
||||
|
||||
public boolean getDebug() {
|
||||
return debug;
|
||||
}
|
||||
|
||||
public String getProvider() {
|
||||
return provider;
|
||||
}
|
||||
|
||||
public String getModelType() {
|
||||
return modelType;
|
||||
}
|
||||
|
||||
public String getModelingUnit() {
|
||||
return modelingUnit;
|
||||
}
|
||||
|
||||
public String getBpeVocab() {
|
||||
return bpeVocab;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private OnlineTransducerModelConfig transducer = OnlineTransducerModelConfig.builder().build();
|
||||
private OnlineParaformerModelConfig paraformer = OnlineParaformerModelConfig.builder().build();
|
||||
private OnlineZipformer2CtcModelConfig zipformer2Ctc = OnlineZipformer2CtcModelConfig.builder().build();
|
||||
private OnlineNeMoCtcModelConfig neMoCtc = OnlineNeMoCtcModelConfig.builder().build();
|
||||
private OnlineToneCtcModelConfig toneCtc = OnlineToneCtcModelConfig.builder().build();
|
||||
private String tokens = "";
|
||||
private int numThreads = 1;
|
||||
private boolean debug = false;
|
||||
private String provider = "cpu";
|
||||
private String modelType = "";
|
||||
private String modelingUnit = "cjkchar";
|
||||
private String bpeVocab = "";
|
||||
|
||||
public Builder setTransducer(OnlineTransducerModelConfig transducer) {
|
||||
this.transducer = transducer;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setTokens(String tokens) {
|
||||
this.tokens = tokens;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setModelType(String modelType) {
|
||||
this.modelType = modelType;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setNumThreads(int numThreads) {
|
||||
this.numThreads = numThreads;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setProvider(String provider) {
|
||||
this.provider = provider;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setModelingUnit(String modelingUnit) {
|
||||
this.modelingUnit = modelingUnit;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setBpeVocab(String bpeVocab) {
|
||||
this.bpeVocab = bpeVocab;
|
||||
return this;
|
||||
}
|
||||
|
||||
public OnlineModelConfig build() {
|
||||
return new OnlineModelConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
-30
@@ -1,30 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineNeMoCtcModelConfig {
|
||||
private final String model;
|
||||
|
||||
private OnlineNeMoCtcModelConfig(Builder builder) {
|
||||
this.model = builder.model;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public String getModel() {
|
||||
return model;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private String model = "";
|
||||
|
||||
public Builder setModel(String model) {
|
||||
this.model = model;
|
||||
return this;
|
||||
}
|
||||
|
||||
public OnlineNeMoCtcModelConfig build() {
|
||||
return new OnlineNeMoCtcModelConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
-42
@@ -1,42 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineParaformerModelConfig {
|
||||
private final String encoder;
|
||||
private final String decoder;
|
||||
|
||||
private OnlineParaformerModelConfig(Builder builder) {
|
||||
this.encoder = builder.encoder;
|
||||
this.decoder = builder.decoder;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public String getEncoder() {
|
||||
return encoder;
|
||||
}
|
||||
|
||||
public String getDecoder() {
|
||||
return decoder;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private String encoder = "";
|
||||
private String decoder = "";
|
||||
|
||||
public Builder setEncoder(String encoder) {
|
||||
this.encoder = encoder;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setDecoder(String decoder) {
|
||||
this.decoder = decoder;
|
||||
return this;
|
||||
}
|
||||
|
||||
public OnlineParaformerModelConfig build() {
|
||||
return new OnlineParaformerModelConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -1,35 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineStream {
|
||||
private long ptr;
|
||||
|
||||
public OnlineStream(long ptr) {
|
||||
this.ptr = ptr;
|
||||
}
|
||||
|
||||
public long getPtr() {
|
||||
return ptr;
|
||||
}
|
||||
|
||||
public void acceptWaveform(float[] samples, int sampleRate) {
|
||||
acceptWaveform(ptr, samples, sampleRate);
|
||||
}
|
||||
|
||||
public void release() {
|
||||
if (ptr == 0) {
|
||||
return;
|
||||
}
|
||||
delete(ptr);
|
||||
ptr = 0;
|
||||
}
|
||||
|
||||
@Override
|
||||
protected void finalize() throws Throwable {
|
||||
release();
|
||||
super.finalize();
|
||||
}
|
||||
|
||||
private native void acceptWaveform(long ptr, float[] samples, int sampleRate);
|
||||
|
||||
private native void delete(long ptr);
|
||||
}
|
||||
-30
@@ -1,30 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineToneCtcModelConfig {
|
||||
private final String model;
|
||||
|
||||
private OnlineToneCtcModelConfig(Builder builder) {
|
||||
this.model = builder.model;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public String getModel() {
|
||||
return model;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private String model = "";
|
||||
|
||||
public Builder setModel(String model) {
|
||||
this.model = model;
|
||||
return this;
|
||||
}
|
||||
|
||||
public OnlineToneCtcModelConfig build() {
|
||||
return new OnlineToneCtcModelConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
-54
@@ -1,54 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineTransducerModelConfig {
|
||||
private final String encoder;
|
||||
private final String decoder;
|
||||
private final String joiner;
|
||||
|
||||
private OnlineTransducerModelConfig(Builder builder) {
|
||||
this.encoder = builder.encoder;
|
||||
this.decoder = builder.decoder;
|
||||
this.joiner = builder.joiner;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public String getEncoder() {
|
||||
return encoder;
|
||||
}
|
||||
|
||||
public String getDecoder() {
|
||||
return decoder;
|
||||
}
|
||||
|
||||
public String getJoiner() {
|
||||
return joiner;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private String encoder = "";
|
||||
private String decoder = "";
|
||||
private String joiner = "";
|
||||
|
||||
public Builder setEncoder(String encoder) {
|
||||
this.encoder = encoder;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setDecoder(String decoder) {
|
||||
this.decoder = decoder;
|
||||
return this;
|
||||
}
|
||||
|
||||
public Builder setJoiner(String joiner) {
|
||||
this.joiner = joiner;
|
||||
return this;
|
||||
}
|
||||
|
||||
public OnlineTransducerModelConfig build() {
|
||||
return new OnlineTransducerModelConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
-30
@@ -1,30 +0,0 @@
|
||||
package com.k2fsa.sherpa.onnx;
|
||||
|
||||
public class OnlineZipformer2CtcModelConfig {
|
||||
private final String model;
|
||||
|
||||
private OnlineZipformer2CtcModelConfig(Builder builder) {
|
||||
this.model = builder.model;
|
||||
}
|
||||
|
||||
public static Builder builder() {
|
||||
return new Builder();
|
||||
}
|
||||
|
||||
public String getModel() {
|
||||
return model;
|
||||
}
|
||||
|
||||
public static class Builder {
|
||||
private String model = "";
|
||||
|
||||
public Builder setModel(String model) {
|
||||
this.model = model;
|
||||
return this;
|
||||
}
|
||||
|
||||
public OnlineZipformer2CtcModelConfig build() {
|
||||
return new OnlineZipformer2CtcModelConfig(this);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -195,7 +195,6 @@ class DashboardController {
|
||||
private offFrameMetrics: (() => void) | null = null;
|
||||
private offFirmwareInfo: (() => void) | null = null;
|
||||
private offVoiceStatus: (() => void) | null = null;
|
||||
private offVoiceWakeWord: (() => void) | null = null;
|
||||
private offAndroidNotification: (() => void) | null = null;
|
||||
private lastInput = "waiting...";
|
||||
private lastSys = "none yet";
|
||||
@@ -1011,11 +1010,6 @@ class DashboardController {
|
||||
this.offVoiceStatus = voiceControlBridge.onStatus((state) => {
|
||||
this.appendLog(state.status);
|
||||
});
|
||||
this.offVoiceWakeWord = voiceControlBridge.onWakeWord((keyword) => {
|
||||
void this.handleWakeWord(keyword).catch((error) => {
|
||||
this.appendLog(`wake-word handler failed: ${this.formatError(error)}`);
|
||||
});
|
||||
});
|
||||
|
||||
await mediaControllerBridge.start();
|
||||
await nightscoutBridge.start();
|
||||
@@ -1082,8 +1076,6 @@ class DashboardController {
|
||||
this.offFirmwareInfo = null;
|
||||
this.offVoiceStatus?.();
|
||||
this.offVoiceStatus = null;
|
||||
this.offVoiceWakeWord?.();
|
||||
this.offVoiceWakeWord = null;
|
||||
await mediaControllerBridge.stop().catch(() => {});
|
||||
await nightscoutBridge.stop().catch(() => {});
|
||||
voiceControlBridge.stop();
|
||||
@@ -1182,8 +1174,6 @@ class DashboardController {
|
||||
this.offFirmwareInfo = null;
|
||||
this.offVoiceStatus?.();
|
||||
this.offVoiceStatus = null;
|
||||
this.offVoiceWakeWord?.();
|
||||
this.offVoiceWakeWord = null;
|
||||
|
||||
const communicator = this.communicator;
|
||||
this.communicator = null;
|
||||
@@ -1816,20 +1806,6 @@ class DashboardController {
|
||||
this.updateCompositePreview();
|
||||
}
|
||||
|
||||
private async handleWakeWord(keyword: string): Promise<void> {
|
||||
const normalized = keyword.trim();
|
||||
if (normalized.length === 0) return;
|
||||
this.appendLog(`wake-word detected: ${normalized}`);
|
||||
const woke = shell.wake("sidebar");
|
||||
if (woke) {
|
||||
const ready = await this.ensureEvenHubSessionActive();
|
||||
if (!ready) {
|
||||
this.appendLog("EvenHub wake barrier timed out for phone wakeword");
|
||||
}
|
||||
this.requestShellRender();
|
||||
}
|
||||
}
|
||||
|
||||
private async handleAndroidNotificationPosted(notificationKey: string): Promise<void> {
|
||||
if (!notificationKey) {
|
||||
this.requestShellRender();
|
||||
|
||||
@@ -0,0 +1,176 @@
|
||||
import { Utils } from "@nativescript/core";
|
||||
|
||||
declare const com: any;
|
||||
declare const java: any;
|
||||
|
||||
/**
|
||||
* Download management for the on-device transcription model (Moonshine, via
|
||||
* sherpa-onnx). The model is no longer bundled in the APK; it is fetched on
|
||||
* demand into the same filesDir location that earlier releases copied the
|
||||
* bundled files to, so upgraded installs that already used on-device
|
||||
* transcription need no download. FaceclawVoiceController reads the files
|
||||
* from this directory.
|
||||
*
|
||||
* Mirrors the on-phone assistant model flow in llama.ts, except the model is
|
||||
* three files rather than one; they download sequentially through
|
||||
* FaceclawModelDownloader (resume + pinned sha256 per file).
|
||||
*/
|
||||
|
||||
type AsrModelFile = {
|
||||
name: string;
|
||||
sha256: string;
|
||||
sizeBytes: number;
|
||||
};
|
||||
|
||||
export const ASR_MODEL = {
|
||||
label: "Moonshine (English)",
|
||||
dirName: "sherpa-onnx-moonshine-base-en-quantized-2026-02-27",
|
||||
// Hugging Face mirror of the sherpa-onnx release asset of the same name.
|
||||
// The GitHub release only offers a tar.bz2, which the phone can't unpack;
|
||||
// this mirror serves the same files (sha256-verified) individually.
|
||||
baseUrl:
|
||||
"https://huggingface.co/csukuangfj2/sherpa-onnx-moonshine-base-en-quantized-2026-02-27/resolve/main/",
|
||||
files: [
|
||||
{
|
||||
name: "decoder_model_merged.ort",
|
||||
sha256: "d9d7b333af34bc552580576ddcf248a1c6c839e0d3b43b09afb9376ed009899d",
|
||||
sizeBytes: 109424400,
|
||||
},
|
||||
{
|
||||
name: "encoder_model.ort",
|
||||
sha256: "7c66495948d0d08ec1af454cd4b5514862ae6511e94712a60e6d83eaec8dc8cf",
|
||||
sizeBytes: 31326816,
|
||||
},
|
||||
{
|
||||
name: "tokens.txt",
|
||||
sha256: "2870d843e14c1e187bf1913a521562a63b53933814bd7f2145120468f494a049",
|
||||
sizeBytes: 549350,
|
||||
},
|
||||
] as AsrModelFile[],
|
||||
totalBytes: 141300566,
|
||||
};
|
||||
|
||||
export type AsrModelState = {
|
||||
status: "absent" | "downloading" | "ready";
|
||||
bytesDownloaded: number;
|
||||
totalBytes: number;
|
||||
};
|
||||
|
||||
let downloader: any = null;
|
||||
// Bytes of files already fully downloaded in this run, plus progress within
|
||||
// the file currently downloading; drives the aggregate percentage.
|
||||
let completedBytes = 0;
|
||||
let currentFileBytes = 0;
|
||||
const stateListeners = new Set<(state: AsrModelState) => void>();
|
||||
|
||||
function modelDirPath(): string {
|
||||
const context = Utils.android.getApplicationContext();
|
||||
return `${context.getFilesDir().getAbsolutePath()}/faceclaw-voice-asr/${ASR_MODEL.dirName}`;
|
||||
}
|
||||
|
||||
function filePath(file: AsrModelFile): string {
|
||||
return `${modelDirPath()}/${file.name}`;
|
||||
}
|
||||
|
||||
function isFilePresent(file: AsrModelFile): boolean {
|
||||
try {
|
||||
const javaFile = new java.io.File(filePath(file));
|
||||
return javaFile.exists() && javaFile.length() > 0;
|
||||
} catch {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
export function isAsrModelReady(): boolean {
|
||||
if (!global.isAndroid) return false;
|
||||
return ASR_MODEL.files.every(isFilePresent);
|
||||
}
|
||||
|
||||
export function asrModelState(): AsrModelState {
|
||||
if (downloader) {
|
||||
return {
|
||||
status: "downloading",
|
||||
bytesDownloaded: completedBytes + currentFileBytes,
|
||||
totalBytes: ASR_MODEL.totalBytes,
|
||||
};
|
||||
}
|
||||
return {
|
||||
status: isAsrModelReady() ? "ready" : "absent",
|
||||
bytesDownloaded: 0,
|
||||
totalBytes: ASR_MODEL.totalBytes,
|
||||
};
|
||||
}
|
||||
|
||||
export function onAsrModelStateChanged(listener: (state: AsrModelState) => void): () => void {
|
||||
stateListeners.add(listener);
|
||||
return () => stateListeners.delete(listener);
|
||||
}
|
||||
|
||||
function notifyStateChanged(): void {
|
||||
const state = asrModelState();
|
||||
stateListeners.forEach((listener) => listener(state));
|
||||
}
|
||||
|
||||
export function startAsrModelDownload(): void {
|
||||
if (!global.isAndroid || downloader || isAsrModelReady()) return;
|
||||
completedBytes = ASR_MODEL.files.filter(isFilePresent).reduce((sum, f) => sum + f.sizeBytes, 0);
|
||||
downloadNextFile();
|
||||
notifyStateChanged();
|
||||
}
|
||||
|
||||
function downloadNextFile(): void {
|
||||
const nextFile = ASR_MODEL.files.find((file) => !isFilePresent(file));
|
||||
if (!nextFile) {
|
||||
downloader = null;
|
||||
notifyStateChanged();
|
||||
return;
|
||||
}
|
||||
currentFileBytes = 0;
|
||||
const listener = new com.faceclaw.app.FaceclawModelDownloaderListener({
|
||||
onProgress: (bytes: number, _total: number) => {
|
||||
currentFileBytes = Number(bytes);
|
||||
notifyStateChanged();
|
||||
},
|
||||
onDone: () => {
|
||||
completedBytes += nextFile.sizeBytes;
|
||||
currentFileBytes = 0;
|
||||
downloadNextFile();
|
||||
},
|
||||
onError: (message: string) => {
|
||||
console.error(`Voice model download failed (${nextFile.name}): ${message}`);
|
||||
downloader = null;
|
||||
notifyStateChanged();
|
||||
},
|
||||
});
|
||||
downloader = new com.faceclaw.app.FaceclawModelDownloader(
|
||||
`${ASR_MODEL.baseUrl}${nextFile.name}`,
|
||||
filePath(nextFile),
|
||||
nextFile.sha256,
|
||||
nextFile.sizeBytes,
|
||||
listener,
|
||||
);
|
||||
downloader.start();
|
||||
}
|
||||
|
||||
/** Stops the download; already-fetched bytes are kept and resumed next time. */
|
||||
export function cancelAsrModelDownload(): void {
|
||||
if (!downloader) return;
|
||||
downloader.cancel();
|
||||
downloader = null;
|
||||
notifyStateChanged();
|
||||
}
|
||||
|
||||
export function deleteAsrModel(): void {
|
||||
if (!global.isAndroid) return;
|
||||
cancelAsrModelDownload();
|
||||
try {
|
||||
for (const file of ASR_MODEL.files) {
|
||||
new java.io.File(filePath(file)).delete();
|
||||
new java.io.File(`${filePath(file)}.part`).delete();
|
||||
}
|
||||
new java.io.File(modelDirPath()).delete();
|
||||
} catch (error) {
|
||||
console.error(`Voice model delete failed: ${String(error)}`);
|
||||
}
|
||||
notifyStateChanged();
|
||||
}
|
||||
@@ -45,7 +45,6 @@ export type RawPcmListener = (pcm: Uint8Array) => void;
|
||||
|
||||
export class FaceclawVoiceControlBridge {
|
||||
private readonly statusListeners = new Set<(state: VoiceControlState) => void>();
|
||||
private readonly wakeWordListeners = new Set<(keyword: string) => void>();
|
||||
private readonly transcriptListeners = new Set<(event: VoiceTranscriptEvent) => void>();
|
||||
private readonly speechEndListeners = new Set<() => void>();
|
||||
private controller: any | null = null;
|
||||
@@ -72,11 +71,6 @@ export class FaceclawVoiceControlBridge {
|
||||
return () => this.statusListeners.delete(listener);
|
||||
}
|
||||
|
||||
onWakeWord(listener: (keyword: string) => void): () => void {
|
||||
this.wakeWordListeners.add(listener);
|
||||
return () => this.wakeWordListeners.delete(listener);
|
||||
}
|
||||
|
||||
onTranscript(listener: (event: VoiceTranscriptEvent) => void): () => void {
|
||||
this.transcriptListeners.add(listener);
|
||||
return () => this.transcriptListeners.delete(listener);
|
||||
@@ -266,11 +260,6 @@ export class FaceclawVoiceControlBridge {
|
||||
if (this.rawActive) return;
|
||||
this.setStatus(String(status));
|
||||
},
|
||||
onWakeWord: (keyword: string) => {
|
||||
for (const listener of this.wakeWordListeners) {
|
||||
listener(String(keyword));
|
||||
}
|
||||
},
|
||||
onTranscript: (text: string, isFinal: boolean) => {
|
||||
this.emitTranscript(String(text), Boolean(isFinal));
|
||||
},
|
||||
|
||||
@@ -319,7 +319,7 @@ export const voiceProviderSetting = new ConfigSettingEnum<VoiceProvider>({
|
||||
if (value === "soniox") return sonioxApiKeySetting.get().trim().length === 0;
|
||||
return false;
|
||||
},
|
||||
description: "Speech-to-text engine for voice input. ElevenLabs, Whisper, and Soniox are cloud services that need an API key, with significantly better accuracy than on-device transcription.",
|
||||
description: "Speech-to-text engine for voice input. ElevenLabs, Whisper, and Soniox are cloud services that need an API key, with significantly better accuracy than on-device transcription. On-device transcription needs the voice model downloaded (below).",
|
||||
});
|
||||
|
||||
const wakeWordActionLabels: Record<WakeWordAction, string> = {
|
||||
|
||||
@@ -12,6 +12,14 @@ import {
|
||||
onLocalModelStateChanged,
|
||||
startLocalModelDownload,
|
||||
} from "../../native/llama";
|
||||
import {
|
||||
ASR_MODEL,
|
||||
asrModelState,
|
||||
cancelAsrModelDownload,
|
||||
deleteAsrModel,
|
||||
onAsrModelStateChanged,
|
||||
startAsrModelDownload,
|
||||
} from "../../native/asr-model";
|
||||
import { TextViewerLayer } from "../../apps/files/text-viewer";
|
||||
import type { LayerContext } from "../layers";
|
||||
import { drawRightValueMenuItem, openModalMenu, type MenuItem } from "../menu";
|
||||
@@ -87,6 +95,7 @@ function settingsSections(): SettingsSection[] {
|
||||
items: [
|
||||
enumSettingMenuItem(wakeWordActionSetting),
|
||||
enumSettingMenuItem(voiceProviderSetting),
|
||||
asrModelMenuItem(),
|
||||
],
|
||||
},
|
||||
{
|
||||
@@ -240,6 +249,75 @@ function localModelMenuItem(): MenuItem {
|
||||
};
|
||||
}
|
||||
|
||||
const ASR_MODEL_MB = `${Math.round(ASR_MODEL.totalBytes / 1e6)}MB`;
|
||||
|
||||
let asrModelRenderUnsub: (() => void) | null = null;
|
||||
|
||||
function watchAsrModelDownload(ctx: LayerContext): void {
|
||||
asrModelRenderUnsub?.();
|
||||
asrModelRenderUnsub = onAsrModelStateChanged((state) => {
|
||||
ctx.actions.requestRender();
|
||||
if (state.status !== "downloading") {
|
||||
asrModelRenderUnsub?.();
|
||||
asrModelRenderUnsub = null;
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
function asrModelStatusText(): string {
|
||||
const state = asrModelState();
|
||||
if (state.status === "ready") return "downloaded";
|
||||
if (state.status === "downloading") {
|
||||
const pct = state.totalBytes > 0 ? Math.floor((state.bytesDownloaded / state.totalBytes) * 100) : 0;
|
||||
return `${pct}% of ${ASR_MODEL_MB}`;
|
||||
}
|
||||
return "not downloaded";
|
||||
}
|
||||
|
||||
/** Download/cancel/delete management for the on-device transcription model. */
|
||||
function asrModelMenuItem(): MenuItem {
|
||||
return {
|
||||
label: "On-device voice model",
|
||||
description:
|
||||
`${ASR_MODEL.label} (${ASR_MODEL_MB} download). ` +
|
||||
"Transcribes voice input on the phone itself, with no API key or cloud service. " +
|
||||
"Required for the On-device transcription provider; the cloud providers work without it. " +
|
||||
"An interrupted download resumes where it left off.",
|
||||
onSelect: (ctx) => {
|
||||
const state = asrModelState();
|
||||
const action: MenuItem =
|
||||
state.status === "downloading"
|
||||
? {
|
||||
label: "Cancel download",
|
||||
onSelect: (innerCtx) => {
|
||||
cancelAsrModelDownload();
|
||||
innerCtx.stack.pop();
|
||||
},
|
||||
}
|
||||
: state.status === "ready"
|
||||
? {
|
||||
label: "Delete model",
|
||||
onSelect: (innerCtx) => {
|
||||
deleteAsrModel();
|
||||
innerCtx.stack.pop();
|
||||
},
|
||||
}
|
||||
: {
|
||||
label: `Download (${ASR_MODEL_MB})`,
|
||||
onSelect: (innerCtx) => {
|
||||
startAsrModelDownload();
|
||||
watchAsrModelDownload(innerCtx);
|
||||
innerCtx.stack.pop();
|
||||
},
|
||||
};
|
||||
openModalMenu(ctx, "On-device voice model", [action], 0);
|
||||
},
|
||||
render: ({ image, x, y, width }) => {
|
||||
drawRightValueMenuItem(image, getDefaultSmallFont(), x, y, width, "On-device voice model", asrModelStatusText());
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
/** A row that opens one of the project docs (copied into the bundle under
|
||||
* about/ by webpack.config.js) in the paged text viewer. */
|
||||
function bundledDocMenuItem(fileName: string, label: string): MenuItem {
|
||||
|
||||
Reference in new issue
Block a user