apps/talk.c

   1 /***************************************************************************
   2  *             __________               __   ___.
   3  *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
   4  *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
   5  *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
   6  *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
   7  *                     \/            \/     \/    \/            \/
   8  * $Id$
   9  *
  10  * Copyright (C) 2004 Jörg Hohensohn
  11  *
  12  * This module collects the Talkbox and voice UI functions.
  13  * (Talkbox reads directory names from mp3 clips called thumbnails,
  14  *  the voice UI lets menus and screens "talk" from a voicefile in memory.
  15  *
  16  * This program is free software; you can redistribute it and/or
  17  * modify it under the terms of the GNU General Public License
  18  * as published by the Free Software Foundation; either version 2
  19  * of the License, or (at your option) any later version.
  20  *
  21  * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
  22  * KIND, either express or implied.
  23  *
  24  ****************************************************************************/
  25
  26 #include <stdio.h>
  27 #include <stddef.h>
  28 #include "string-extra.h"
  29 #include "file.h"
  30 #include "buffer.h"
  31 #include "system.h"
  32 #include "kernel.h"
  33 #include "settings.h"
  34 #include "settings_list.h"
  35 #include "mp3_playback.h"
  36 #include "audio.h"
  37 #include "lang.h"
  38 #include "talk.h"
  39 #include "metadata.h"
  40 /*#define LOGF_ENABLE*/
  41 #include "logf.h"
  42 #include "bitswap.h"
  43 #include "structec.h"
  44 #include "debug.h"
  45
  46
  47 /* Memory layout varies between targets because the
  48    Archos (MASCODEC) devices cannot mix voice and audio playback
  49
  50              MASCODEC  | MASCODEC  | SWCODEC
  51              (playing) | (stopped) |
  52     voicebuf-----------+-----------+------------
  53               audio    | voice     | thumbnail
  54                        |-----------|------------
  55                        | thumbnail | voice
  56                        |           |------------
  57                        |           | filebuf
  58                        |           |------------
  59                        |           | audio
  60   voicebufend----------+-----------+------------
  61
  62   SWCODEC allocates dedicated buffers, MASCODEC reuses audiobuf. */
  63
  64
  65 /***************** Constants *****************/
  66
  67 #define QUEUE_SIZE 64 /* must be a power of two */
  68 #define QUEUE_MASK (QUEUE_SIZE-1)
  69 const char* const dir_thumbnail_name = "_dirname.talk";
  70 const char* const file_thumbnail_ext = ".talk";
  71
  72 /***************** Functional Macros *****************/
  73
  74 #define QUEUE_LEVEL ((queue_write - queue_read) & QUEUE_MASK)
  75
  76 #define LOADED_MASK 0x80000000 /* MSB */
  77
  78 #if CONFIG_CODEC == SWCODEC
  79 #define MAX_THUMBNAIL_BUFSIZE 0x10000
  80 #endif
  81
  82 /***************** Data types *****************/
  83
  84 struct clip_entry /* one entry of the index table */
  85 {
  86     int offset; /* offset from start of voicefile file */
  87     int size; /* size of the clip */
  88 };
  89
  90 struct voicefile /* file format of our voice file */
  91 {
  92     int version; /* version of the voicefile */
  93     int target_id; /* the rockbox target the file was made for */
  94     int table;   /* offset to index table, (=header size) */
  95     int id1_max; /* number of "normal" clips contained in above index */
  96     int id2_max; /* number of "voice only" clips contained in above index */
  97     struct clip_entry index[]; /* followed by the index tables */
  98     /* and finally the mp3 clips, not visible here, bitswapped
  99        for SH based players */
 100 };
 101
 102 struct queue_entry /* one entry of the internal queue */
 103 {
 104     unsigned char* buf;
 105     long len;
 106 };
 107
 108
 109 /***************** Globals *****************/
 110
 111 #if CONFIG_STORAGE & STORAGE_MMC
 112 /* The MMC storage on the Ondios is slow enough that we want to buffer the
 113  * talk clips only when they are needed */
 114 #   define TALK_PROGRESSIVE_LOAD
 115 #elif CONFIG_CODEC == SWCODEC && MEMORYSIZE <= 2
 116 /* The entire voice file wouldn't fit in memory together with codecs, so we
 117  * load clips each time they are accessed */
 118 #   define TALK_PARTIAL_LOAD
 119 #endif
 120
 121 #ifdef TALK_PARTIAL_LOAD
 122 static unsigned char *clip_buffer;
 123 static long           max_clipsize; /* size of the biggest clip */
 124 static long           buffered_id[QUEUE_SIZE];  /* IDs of the talk clips */
 125 static uint8_t        clip_age[QUEUE_SIZE];
 126 #if QUEUE_SIZE > 255
 127 #   error clip_age[] type too small
 128 #endif
 129 #endif
 130
 131 static char* voicebuf; /* root pointer to our buffer */
 132 static unsigned char* p_thumbnail = NULL; /* buffer for thumbnails */
 133 /* Multiple thumbnails can be loaded back-to-back in this buffer. */
 134 static volatile int thumbnail_buf_used SHAREDBSS_ATTR; /* length of data in
 135                                                           thumbnail buffer */
 136 static long size_for_thumbnail; /* total thumbnail buffer size */
 137 static struct voicefile* p_voicefile; /* loaded voicefile */
 138 static bool has_voicefile; /* a voicefile file is present */
 139 static bool need_shutup; /* is there possibly any voice playing to be shutup */
 140 static struct queue_entry queue[QUEUE_SIZE]; /* queue of scheduled clips */
 141 static bool force_enqueue_next; /* enqueue next utterance even if enqueue is false */
 142 static int queue_write; /* write index of queue, by application */
 143 static int queue_read; /* read index of queue, by ISR context */
 144 #if CONFIG_CODEC == SWCODEC
 145 /* protects queue_read, queue_write and thumbnail_buf_used */
 146 static struct mutex queue_mutex SHAREDBSS_ATTR;
 147 #define talk_queue_lock() ({ mutex_lock(&queue_mutex); })
 148 #define talk_queue_unlock() ({ mutex_unlock(&queue_mutex); })
 149 #else
 150 #define talk_queue_lock() ({ })
 151 #define talk_queue_unlock() ({ })
 152 #endif /* CONFIG_CODEC */
 153 static int sent; /* how many bytes handed over to playback, owned by ISR */
 154 static unsigned char curr_hd[3]; /* current frame header, for re-sync */
 155 static int filehandle = -1; /* global, so we can keep the file open if needed */
 156 static unsigned char* p_silence; /* VOICE_PAUSE clip, used for termination */
 157 static long silence_len; /* length of the VOICE_PAUSE clip */
 158 static unsigned char* p_lastclip; /* address of latest clip, for silence add */
 159 static unsigned long voicefile_size = 0; /* size of the loaded voice file */
 160 static unsigned char last_lang[MAX_FILENAME+1]; /* name of last used lang file (in talk_init) */
 161 static bool talk_initialized; /* true if talk_init has been called */
 162 static int talk_temp_disable_count; /* if positive, temporarily disable voice UI (not saved) */
 163
 164
 165 /***************** Private implementation *****************/
 166
 167 static int open_voicefile(void)
 168 {
 169     char buf[64];
 170     char* p_lang = "english"; /* default */
 171
 172     if ( global_settings.lang_file[0] &&
 173          global_settings.lang_file[0] != 0xff )
 174     {   /* try to open the voice file of the selected language */
 175         p_lang = (char *)global_settings.lang_file;
 176     }
 177
 178     snprintf(buf, sizeof(buf), LANG_DIR "/%s.voice", p_lang);
 179
 180     return open(buf, O_RDONLY);
 181 }
 182
 183
 184 /* fetch a clip from the voice file */
 185 static unsigned char* get_clip(long id, long* p_size)
 186 {
 187     long clipsize;
 188     unsigned char* clipbuf;
 189
 190     if (id > VOICEONLY_DELIMITER)
 191     {   /* voice-only entries use the second part of the table */
 192         id -= VOICEONLY_DELIMITER + 1;
 193         if (id >= p_voicefile->id2_max)
 194             return NULL; /* must be newer than we have */
 195         id += p_voicefile->id1_max; /* table 2 is behind table 1 */
 196     }
 197     else
 198     {   /* normal use of the first table */
 199         if (id >= p_voicefile->id1_max)
 200             return NULL; /* must be newer than we have */
 201     }
 202
 203     clipsize = p_voicefile->index[id].size;
 204     if (clipsize == 0) /* clip not included in voicefile */
 205         return NULL;
 206
 207 #ifndef TALK_PARTIAL_LOAD
 208     clipbuf = (unsigned char *) p_voicefile + p_voicefile->index[id].offset;
 209 #endif
 210
 211 #if defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD)
 212     if (!(clipsize & LOADED_MASK))
 213     {   /* clip needs loading */
 214 #ifdef TALK_PARTIAL_LOAD
 215         int idx = 0;
 216         if (id == VOICE_PAUSE) {
 217             idx = QUEUE_SIZE;   /* we keep VOICE_PAUSE loaded */
 218         } else {
 219             int oldest = 0, i;
 220             for(i=0; i<QUEUE_SIZE; i++) {
 221                 if (buffered_id[i] < 0) {
 222                     /* found a free entry, that means the buffer isn't
 223                      * full yet. */
 224                     idx = i;
 225                     break;
 226                 }
 227
 228                 /* find the oldest clip */
 229                 if(clip_age[i] > oldest) {
 230                     idx = i;
 231                     oldest = clip_age[i];
 232                 }
 233
 234                 /* increment age of each loaded clip */
 235                 clip_age[i]++;
 236             }
 237             clip_age[idx] = 0; /* reset clip's age */
 238         }
 239         clipbuf = clip_buffer + idx * max_clipsize;
 240 #endif
 241
 242         lseek(filehandle, p_voicefile->index[id].offset, SEEK_SET);
 243         if (read(filehandle, clipbuf, clipsize) != clipsize)
 244             return NULL; /* read error */
 245
 246         p_voicefile->index[id].size |= LOADED_MASK; /* mark as loaded */
 247
 248 #ifdef TALK_PARTIAL_LOAD
 249         if (id != VOICE_PAUSE) {
 250             if (buffered_id[idx] >= 0) {
 251                 /* mark previously loaded clip as unloaded */
 252                 p_voicefile->index[buffered_id[idx]].size &= ~LOADED_MASK;
 253             }
 254             buffered_id[idx] = id;
 255         }
 256 #endif
 257     }
 258     else
 259     {   /* clip is in memory already */
 260 #ifdef TALK_PARTIAL_LOAD
 261         /* Find where it was loaded */
 262         clipbuf = clip_buffer;
 263         if (id == VOICE_PAUSE) {
 264             clipbuf += QUEUE_SIZE * max_clipsize;
 265         } else {
 266             int idx;
 267             for (idx=0; idx<QUEUE_SIZE; idx++)
 268                 if (buffered_id[idx] == id) {
 269                     clipbuf += idx * max_clipsize;
 270                     clip_age[idx] = 0; /* reset clip's age */
 271                     break;
 272                 }
 273         }
 274 #endif
 275         clipsize &= ~LOADED_MASK; /* without the extra bit gives true size */
 276     }
 277 #endif /* defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD) */
 278
 279     *p_size = clipsize;
 280     return clipbuf;
 281 }
 282
 283
 284 /* load the voice file into the mp3 buffer */
 285 static void load_voicefile(bool probe, char* buf, size_t bufsize)
 286 {
 287     union voicebuf {
 288         unsigned char*    buf;
 289         struct voicefile* file;
 290     };
 291     union voicebuf voicebuf;
 292
 293     int load_size, alloc_size;
 294     int got_size;
 295 #ifndef TALK_PARTIAL_LOAD
 296     size_t file_size;
 297 #endif
 298 #ifdef ROCKBOX_LITTLE_ENDIAN
 299     int i;
 300 #endif
 301
 302     if (!probe)
 303         filehandle = open_voicefile();
 304     if (filehandle < 0) /* failed to open */
 305         goto load_err;
 306
 307     voicebuf.buf = buf;
 308     if (!voicebuf.buf)
 309         goto load_err;
 310
 311 #ifndef TALK_PARTIAL_LOAD
 312     file_size = filesize(filehandle);
 313     if (file_size > bufsize) /* won't fit? */
 314         goto load_err;
 315 #endif
 316
 317 #if defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD)
 318     /* load only the header for now */
 319     load_size = sizeof(struct voicefile);
 320 #else /* load the full file */
 321     load_size = file_size;
 322 #endif
 323
 324 #ifdef TALK_PARTIAL_LOAD
 325     if (load_size > bufsize) /* won't fit? */
 326         goto load_err;
 327 #endif
 328
 329     got_size = read(filehandle, voicebuf.buf, load_size);
 330     if (got_size != load_size /* failure */)
 331         goto load_err;
 332
 333     alloc_size = load_size;
 334
 335 #ifdef ROCKBOX_LITTLE_ENDIAN
 336     logf("Byte swapping voice file");
 337     structec_convert(voicebuf.buf, "lllll", 1, true);
 338 #endif
 339
 340     /* format check */
 341     if (voicebuf.file->table == sizeof(struct voicefile))
 342     {
 343         p_voicefile = voicebuf.file;
 344
 345         if (p_voicefile->version != VOICE_VERSION ||
 346             p_voicefile->target_id != TARGET_ID)
 347         {
 348             logf("Incompatible voice file");
 349             goto load_err;
 350         }
 351 #if CONFIG_CODEC != SWCODEC
 352         /* MASCODEC: now use audiobuf for voice then thumbnail */
 353         p_thumbnail = voicebuf.buf + file_size;
 354         p_thumbnail += (long)p_thumbnail % 2; /* 16-bit align */
 355         size_for_thumbnail = voicebuf.buf + bufsize - p_thumbnail;
 356 #endif
 357     }
 358     else
 359         goto load_err;
 360
 361 #if defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD)
 362     /* load the index table, now that we know its size from the header */
 363     load_size = (p_voicefile->id1_max + p_voicefile->id2_max)
 364                 * sizeof(struct clip_entry);
 365
 366 #ifdef TALK_PARTIAL_LOAD
 367     if (load_size > bufsize) /* won't fit? */
 368         goto load_err;
 369 #endif
 370
 371     got_size = read(filehandle, &p_voicefile->index[0], load_size);
 372     if (got_size != load_size) /* read error */
 373         goto load_err;
 374
 375     alloc_size += load_size;
 376 #else
 377     close(filehandle);
 378     filehandle = -1;
 379 #endif /* defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD) */
 380
 381 #ifdef ROCKBOX_LITTLE_ENDIAN
 382     for (i = 0; i < p_voicefile->id1_max + p_voicefile->id2_max; i++)
 383         structec_convert(&p_voicefile->index[i], "ll", 1, true);
 384 #endif
 385
 386 #ifdef TALK_PARTIAL_LOAD
 387     clip_buffer = (unsigned char *) p_voicefile + p_voicefile->table;
 388     unsigned clips = p_voicefile->id1_max + p_voicefile->id2_max;
 389     clip_buffer += clips * sizeof(struct clip_entry); /* skip index */
 390 #endif
 391     if (!probe) {
 392         /* make sure to have the silence clip, if available */
 393         p_silence = get_clip(VOICE_PAUSE, &silence_len);
 394     }
 395
 396 #ifdef TALK_PARTIAL_LOAD
 397     alloc_size += silence_len + QUEUE_SIZE;
 398 #endif
 399     if ((size_t)alloc_size > bufsize)
 400         goto load_err;
 401     return;
 402
 403 load_err:
 404     p_voicefile = NULL;
 405     has_voicefile = false; /* don't try again */
 406     if (filehandle >= 0)
 407     {
 408         close(filehandle);
 409         filehandle = -1;
 410     }
 411     return;
 412 }
 413
 414
 415 /* called in ISR context if mp3 data got consumed */
 416 static void mp3_callback(unsigned char** start, size_t* size)
 417 {
 418     queue[queue_read].len -= sent; /* we completed this */
 419     queue[queue_read].buf += sent;
 420
 421     if (queue[queue_read].len > 0) /* current clip not finished? */
 422     {   /* feed the next 64K-1 chunk */
 423 #if CONFIG_CODEC != SWCODEC
 424         sent = MIN(queue[queue_read].len, 0xFFFF);
 425 #else
 426         sent = queue[queue_read].len;
 427 #endif
 428         *start = queue[queue_read].buf;
 429         *size = sent;
 430         return;
 431     }
 432     talk_queue_lock();
 433     if(p_thumbnail
 434        && queue[queue_read].buf == p_thumbnail +thumbnail_buf_used)
 435         thumbnail_buf_used = 0;
 436     if (sent > 0) /* go to next entry */
 437     {
 438         queue_read = (queue_read + 1) & QUEUE_MASK;
 439     }
 440
 441 re_check:
 442
 443     if (QUEUE_LEVEL != 0) /* queue is not empty? */
 444     {   /* start next clip */
 445 #if CONFIG_CODEC != SWCODEC
 446         sent = MIN(queue[queue_read].len, 0xFFFF);
 447 #else
 448         sent = queue[queue_read].len;
 449 #endif
 450         *start = p_lastclip = queue[queue_read].buf;
 451         *size = sent;
 452         curr_hd[0] = p_lastclip[1];
 453         curr_hd[1] = p_lastclip[2];
 454         curr_hd[2] = p_lastclip[3];
 455     }
 456     else if (p_silence != NULL             /* silence clip available */
 457              && p_lastclip != p_silence    /* previous clip wasn't silence */
 458              && !(p_lastclip >= p_thumbnail /* ..or thumbnail */
 459                   && p_lastclip < p_thumbnail +size_for_thumbnail))
 460     {   /* add silence clip when queue runs empty playing a voice clip */
 461         queue[queue_write].buf = p_silence;
 462         queue[queue_write].len = silence_len;
 463         queue_write = (queue_write + 1) & QUEUE_MASK;
 464
 465         goto re_check;
 466     }
 467     else
 468     {
 469         *size = 0; /* end of data */
 470     }
 471     talk_queue_unlock();
 472 }
 473
 474 /***************** Public routines *****************/
 475
 476 /* stop the playback and the pending clips */
 477 void talk_force_shutup(void)
 478 {
 479     /* Most of this is MAS only */
 480 #if CONFIG_CODEC != SWCODEC
 481 #ifdef SIMULATOR
 482     return;
 483 #endif
 484     unsigned char* pos;
 485     unsigned char* search;
 486     unsigned char* end;
 487     if (QUEUE_LEVEL == 0) /* has ended anyway */
 488         return;
 489
 490 #if CONFIG_CPU == SH7034
 491     CHCR3 &= ~0x0001; /* disable the DMA (and therefore the interrupt also) */
 492 #endif /* CONFIG_CPU == SH7034 */
 493     /* search next frame boundary and continue up to there */
 494     pos = search = mp3_get_pos();
 495     end = queue[queue_read].buf + queue[queue_read].len;
 496
 497     if (pos >= queue[queue_read].buf
 498         && pos <= end) /* really our clip? */
 499     { /* (for strange reasons this isn't nesessarily the case) */
 500         /* find the next frame boundary */
 501         while (search < end) /* search the remaining data */
 502         {
 503             if (*search++ != 0xFF) /* quick search for frame sync byte */
 504                 continue; /* (this does the majority of the job) */
 505
 506             /* look at the (bitswapped) rest of header candidate */
 507             if (search[0] == curr_hd[0] /* do the quicker checks first */
 508              && search[2] == curr_hd[2]
 509              && (search[1] & 0x30) == (curr_hd[1] & 0x30)) /* sample rate */
 510             {
 511                 search--; /* back to the sync byte */
 512                 break; /* From looking at it, this is our header. */
 513             }
 514         }
 515
 516         if (search-pos)
 517         {   /* play old data until the frame end, to keep the MAS in sync */
 518             sent = search-pos;
 519
 520             queue_write = (queue_read + 1) & QUEUE_MASK; /* will be empty after next callback */
 521             queue[queue_read].len = sent; /* current one ends after this */
 522
 523 #if CONFIG_CPU == SH7034
 524             DTCR3 = sent; /* let the DMA finish this frame */
 525             CHCR3 |= 0x0001; /* re-enable DMA */
 526 #endif /* CONFIG_CPU == SH7034 */
 527             thumbnail_buf_used = 0;
 528             return;
 529         }
 530     }
 531 #endif /* CONFIG_CODEC != SWCODEC */
 532
 533     /* Either SWCODEC, or MAS had nothing to do (was frame boundary or not our clip) */
 534     mp3_play_stop();
 535     talk_queue_lock();
 536     queue_write = queue_read = 0; /* reset the queue */
 537     thumbnail_buf_used = 0;
 538     talk_queue_unlock();
 539     need_shutup = false;
 540 }
 541
 542 /* Shutup the voice, except if force_enqueue_next is set. */
 543 void talk_shutup(void)
 544 {
 545     if (need_shutup && !force_enqueue_next)
 546         talk_force_shutup();
 547 }
 548
 549 /* schedule a clip, at the end or discard the existing queue */
 550 static void queue_clip(unsigned char* buf, long size, bool enqueue)
 551 {
 552     int queue_level;
 553
 554     if (!enqueue)
 555         talk_shutup(); /* cut off all the pending stuff */
 556     /* Something is being enqueued, force_enqueue_next override is no
 557        longer in effect. */
 558     force_enqueue_next = false;
 559
 560     if (!size)
 561         return; /* safety check */
 562 #if CONFIG_CPU == SH7034
 563     /* disable the DMA temporarily, to be safe of race condition */
 564     CHCR3 &= ~0x0001;
 565 #endif
 566     talk_queue_lock();
 567     queue_level = QUEUE_LEVEL; /* check old level */
 568
 569     if (queue_level < QUEUE_SIZE - 1) /* space left? */
 570     {
 571         queue[queue_write].buf = buf; /* populate an entry */
 572         queue[queue_write].len = size;
 573         queue_write = (queue_write + 1) & QUEUE_MASK;
 574     }
 575     talk_queue_unlock();
 576
 577     if (queue_level == 0)
 578     {   /* queue was empty, we have to do the initial start */
 579         p_lastclip = buf;
 580 #if CONFIG_CODEC != SWCODEC
 581         sent = MIN(size, 0xFFFF); /* DMA can do no more */
 582 #else
 583         sent = size;
 584 #endif
 585         mp3_play_data(buf, sent, mp3_callback);
 586         curr_hd[0] = buf[1];
 587         curr_hd[1] = buf[2];
 588         curr_hd[2] = buf[3];
 589         mp3_play_pause(true); /* kickoff audio */
 590     }
 591     else
 592     {
 593 #if CONFIG_CPU == SH7034
 594         CHCR3 |= 0x0001; /* re-enable DMA */
 595 #endif
 596     }
 597
 598     need_shutup = true;
 599
 600     return;
 601 }
 602
 603
 604 static void alloc_thumbnail_buf(void)
 605 {
 606 #if CONFIG_CODEC == SWCODEC
 607     /* Allocate a dedicated thumbnail buffer - once */
 608     if (p_thumbnail == NULL)
 609     {
 610         size_for_thumbnail = buffer_available();
 611         if (size_for_thumbnail > MAX_THUMBNAIL_BUFSIZE)
 612             size_for_thumbnail = MAX_THUMBNAIL_BUFSIZE;
 613         p_thumbnail = buffer_alloc(size_for_thumbnail);
 614     }
 615 #else
 616     /* use the audio buffer now, need to release before loading a voice */
 617     p_thumbnail = voicebuf;
 618 #endif
 619     thumbnail_buf_used = 0;
 620 }
 621
 622 /* common code for talk_init() and talk_buffer_steal() */
 623 static void reset_state(void)
 624 {
 625     queue_write = queue_read = 0; /* reset the queue */
 626     p_voicefile = NULL; /* indicate no voicefile (trashed) */
 627     p_thumbnail = NULL;
 628
 629 #ifdef TALK_PARTIAL_LOAD
 630     int i;
 631     for(i=0; i<QUEUE_SIZE; i++)
 632         buffered_id[i] = -1;
 633 #endif
 634
 635     p_silence = NULL; /* pause clip not accessible */
 636     voicebuf = NULL;
 637 }
 638
 639
 640 /***************** Public implementation *****************/
 641
 642 void talk_init(void)
 643 {
 644     talk_temp_disable_count = 0;
 645     if (talk_initialized && !strcasecmp(last_lang, global_settings.lang_file))
 646     {
 647         /* not a new file, nothing to do */
 648         return;
 649     }
 650
 651 #if defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD)
 652     if (filehandle >= 0)
 653     {
 654         close(filehandle);
 655         filehandle = -1;
 656     }
 657 #endif
 658
 659 #if CONFIG_CODEC == SWCODEC
 660     if(!talk_initialized)
 661         mutex_init(&queue_mutex);
 662 #endif /* CONFIG_CODEC == SWCODEC */
 663
 664     talk_initialized = true;
 665     strlcpy((char *)last_lang, (char *)global_settings.lang_file,
 666         MAX_FILENAME);
 667
 668     filehandle = open_voicefile();
 669     if (filehandle < 0) {
 670         has_voicefile = false;
 671         voicefile_size = 0;
 672         return;
 673     }
 674
 675     voicefile_size = filesize(filehandle);
 676
 677 #if CONFIG_CODEC == SWCODEC
 678     audio_get_buffer(false, NULL);  /* Must tell audio to reinitialize */
 679 #endif
 680     reset_state(); /* use this for most of our inits */
 681
 682 #ifdef TALK_PARTIAL_LOAD
 683     size_t bufsize;
 684     char* buf = plugin_get_buffer(&bufsize);
 685     /* we won't load the full file, we only need the index */
 686     load_voicefile(true, buf, bufsize);
 687     if (!p_voicefile)
 688         return;
 689
 690     unsigned clips = p_voicefile->id1_max + p_voicefile->id2_max;
 691     unsigned i;
 692     int silence_size = 0;
 693
 694     for(i=0; i<clips; i++) {
 695         int size = p_voicefile->index[i].size;
 696         if (size > max_clipsize)
 697             max_clipsize = size;
 698         if (i == VOICE_PAUSE)
 699             silence_size = size;
 700     }
 701
 702     voicefile_size = p_voicefile->table + clips * sizeof(struct clip_entry);
 703     voicefile_size += max_clipsize * QUEUE_SIZE + silence_size;
 704     p_voicefile = NULL; /* Don't pretend we can load talk clips just yet */
 705 #endif
 706
 707
 708     /* test if we can open and if it fits in the audiobuffer */
 709     size_t audiobufsz = buffer_available();
 710     if (voicefile_size <= audiobufsz) {
 711         has_voicefile = true;
 712     } else {
 713         has_voicefile = false;
 714         voicefile_size = 0;
 715     }
 716
 717     alloc_thumbnail_buf();
 718     close(filehandle); /* close again, this was just to detect presence */
 719     filehandle = -1;
 720 }
 721
 722 #if CONFIG_CODEC == SWCODEC
 723 /* return if a voice codec is required or not */
 724 bool talk_voice_required(void)
 725 {
 726     return (voicefile_size != 0) /* Voice file is available */
 727         || (global_settings.talk_dir_clip)  /* Thumbnail clips are required */
 728         || (global_settings.talk_file_clip);
 729 }
 730 #endif
 731
 732 /* return size of voice file */
 733 int talk_get_buffer(void)
 734 {
 735     return voicefile_size;
 736 }
 737
 738 size_t talkbuf_init(char *bufstart)
 739 {
 740     if (bufstart)
 741         voicebuf = bufstart;
 742     else
 743         reset_state();
 744
 745     return voicefile_size;
 746 }
 747
 748 /* somebody else claims the mp3 buffer, e.g. for regular play/record */
 749 void talk_buffer_steal(void)
 750 {
 751 #if CONFIG_CODEC != SWCODEC
 752     mp3_play_stop();
 753 #endif
 754 #if defined(TALK_PROGRESSIVE_LOAD) || defined(TALK_PARTIAL_LOAD)
 755     if (filehandle >= 0)
 756     {
 757         close(filehandle);
 758         filehandle = -1;
 759     }
 760 #endif
 761     reset_state();
 762 }
 763
 764
 765 /* play a voice ID from voicefile */
 766 int talk_id(int32_t id, bool enqueue)
 767 {
 768     long clipsize;
 769     unsigned char* clipbuf;
 770     int32_t unit;
 771     int decimals;
 772
 773     if (talk_temp_disable_count > 0)
 774         return -1;  /* talking has been disabled */
 775 #if CONFIG_CODEC != SWCODEC
 776     if (audio_status()) /* busy, buffer in use */
 777         return -1;
 778 #endif
 779
 780     if (p_voicefile == NULL && has_voicefile)
 781         load_voicefile(false, voicebuf, voicefile_size); /* reload needed */
 782
 783     if (p_voicefile == NULL) /* still no voices? */
 784         return -1;
 785
 786     if (id == -1) /* -1 is an indication for silence */
 787         return -1;
 788
 789     decimals = (((uint32_t)id) >> DECIMAL_SHIFT) & 0x7;
 790
 791     /* check if this is a special ID, with a value */
 792     unit = ((uint32_t)id) >> UNIT_SHIFT;
 793     if (unit || decimals)
 794     {   /* sign-extend the value */
 795         id = (uint32_t)id << (32-DECIMAL_SHIFT);
 796         id >>= (32-DECIMAL_SHIFT);
 797
 798         talk_value_decimal(id, unit, decimals, enqueue); /* speak it */
 799         return 0; /* and stop, end of special case */
 800     }
 801
 802     clipbuf = get_clip(id, &clipsize);
 803     if (clipbuf == NULL)
 804         return -1; /* not present */
 805
 806 #ifdef LOGF_ENABLE
 807     if (id > VOICEONLY_DELIMITER)
 808         logf("\ntalk_id: Say voice clip 0x%x\n", id);
 809     else
 810         logf("\ntalk_id: Say '%s'\n", str(id));
 811 #endif
 812
 813     queue_clip(clipbuf, clipsize, enqueue);
 814
 815     return 0;
 816 }
 817 /* Speaks zero or more IDs (from an array). */
 818 int talk_idarray(const long *ids, bool enqueue)
 819 {
 820     int r;
 821     if(!ids)
 822         return 0;
 823     while(*ids != TALK_FINAL_ID)
 824     {
 825         if((r = talk_id(*ids++, enqueue)) <0)
 826             return r;
 827         enqueue = true;
 828     }
 829     return 0;
 830 }
 831
 832 /* Make sure the current utterance is not interrupted by the next one. */
 833 void talk_force_enqueue_next(void)
 834 {
 835     force_enqueue_next = true;
 836 }
 837
 838 /* play a thumbnail from file */
 839 /* Returns size of spoken thumbnail, so >0 means something is spoken,
 840    <=0 means something went wrong. */
 841 static int _talk_file(const char* filename,
 842                       const long *prefix_ids, bool enqueue)
 843 {
 844     int fd;
 845     int size;
 846     int thumb_used;
 847 #if CONFIG_CODEC != SWCODEC
 848     struct mp3entry info;
 849 #endif
 850
 851     if (talk_temp_disable_count > 0)
 852         return -1;  /* talking has been disabled */
 853 #if CONFIG_CODEC != SWCODEC
 854     if (audio_status()) /* busy, buffer in use */
 855         return -1;
 856 #endif
 857
 858     if (p_thumbnail == NULL || size_for_thumbnail <= 0)
 859         alloc_thumbnail_buf();
 860
 861 #if CONFIG_CODEC != SWCODEC
 862     if(mp3info(&info, filename)) /* use this to find real start */
 863     {
 864         return 0; /* failed to open, or invalid */
 865     }
 866 #endif
 867
 868     if (!enqueue)
 869         /* shutup now to free the thumbnail buffer */
 870         talk_shutup();
 871
 872     fd = open(filename, O_RDONLY);
 873     if (fd < 0) /* failed to open */
 874     {
 875         return 0;
 876     }
 877
 878     thumb_used = thumbnail_buf_used;
 879     if(filesize(fd) > size_for_thumbnail -thumb_used)
 880     {   /* Don't play truncated  clips */
 881         close(fd);
 882         return 0;
 883     }
 884
 885 #if CONFIG_CODEC != SWCODEC
 886     lseek(fd, info.first_frame_offset, SEEK_SET); /* behind ID data */
 887 #endif
 888
 889     size = read(fd, p_thumbnail +thumb_used,
 890                 size_for_thumbnail -thumb_used);
 891     close(fd);
 892
 893     /* ToDo: find audio, skip ID headers and trailers */
 894
 895     if (size > 0)    /* Don't play missing clips */
 896     {
 897 #if CONFIG_CODEC != SWCODEC && !defined(SIMULATOR)
 898         bitswap(p_thumbnail, size);
 899 #endif
 900         if(prefix_ids)
 901             /* prefix thumbnail by speaking these ids, but only now
 902                that we know there's actually a thumbnail to be
 903                spoken. */
 904             talk_idarray(prefix_ids, true);
 905         talk_queue_lock();
 906         thumbnail_buf_used = thumb_used +size;
 907         talk_queue_unlock();
 908         queue_clip(p_thumbnail +thumb_used, size, true);
 909     }
 910
 911     return size;
 912 }
 913
 914 int talk_file(const char *root, const char *dir, const char *file,
 915               const char *ext, const long *prefix_ids, bool enqueue)
 916 /* Play a thumbnail file */
 917 {
 918     char buf[MAX_PATH];
 919     /* Does root end with a slash */
 920     char *slash = (root && root[0]
 921                    && root[strlen(root)-1] != '/') ? "/" : "";
 922     snprintf(buf, MAX_PATH, "%s%s%s%s%s%s",
 923              root ? root : "", slash,
 924              dir ? dir : "", dir ? "/" : "",
 925              file ? file : "",
 926              ext ? ext : "");
 927     return _talk_file(buf, prefix_ids, enqueue);
 928 }
 929
 930 static int talk_spell_basename(const char *path,
 931                                const long *prefix_ids, bool enqueue)
 932 {
 933     if(prefix_ids)
 934     {
 935         talk_idarray(prefix_ids, enqueue);
 936         enqueue = true;
 937     }
 938     char buf[MAX_PATH];
 939     /* Spell only the path component after the last slash */
 940     strlcpy(buf, path, sizeof(buf));
 941     if(strlen(buf) >1 && buf[strlen(buf)-1] == '/')
 942         /* strip trailing slash */
 943         buf[strlen(buf)-1] = '\0';
 944     char *ptr = strrchr(buf, '/');
 945     if(ptr && strlen(buf) >1)
 946         ++ptr;
 947     else ptr = buf;
 948     return talk_spell(ptr, enqueue);
 949 }
 950
 951 /* Play a file's .talk thumbnail, fallback to spelling the filename, or
 952    go straight to spelling depending on settings. */
 953 int talk_file_or_spell(const char *dirname, const char *filename,
 954                        const long *prefix_ids, bool enqueue)
 955 {
 956     if (global_settings.talk_file_clip)
 957     {   /* .talk clips enabled */
 958         if(talk_file(dirname, NULL, filename, file_thumbnail_ext,
 959                               prefix_ids, enqueue) >0)
 960             return 0;
 961     }
 962     if (global_settings.talk_file == 2)
 963         /* Either .talk clips are disabled, or as a fallback */
 964         return talk_spell_basename(filename, prefix_ids, enqueue);
 965     return 0;
 966 }
 967
 968 /* Play a directory's .talk thumbnail, fallback to spelling the filename, or
 969    go straight to spelling depending on settings. */
 970 int talk_dir_or_spell(const char* dirname,
 971                       const long *prefix_ids, bool enqueue)
 972 {
 973     if (global_settings.talk_dir_clip)
 974     {   /* .talk clips enabled */
 975         if(talk_file(dirname, NULL, dir_thumbnail_name, NULL,
 976                               prefix_ids, enqueue) >0)
 977             return 0;
 978     }
 979     if (global_settings.talk_dir == 2)
 980         /* Either .talk clips disabled or as a fallback */
 981         return talk_spell_basename(dirname, prefix_ids, enqueue);
 982     return 0;
 983 }
 984
 985 /* say a numeric value, this word ordering works for english,
 986    but not necessarily for other languages (e.g. german) */
 987 int talk_number(long n, bool enqueue)
 988 {
 989     int level = 2; /* mille count */
 990     long mil = 1000000000; /* highest possible "-illion" */
 991
 992     if (talk_temp_disable_count > 0)
 993         return -1;  /* talking has been disabled */
 994 #if CONFIG_CODEC != SWCODEC
 995     if (audio_status()) /* busy, buffer in use */
 996         return -1;
 997 #endif
 998
 999     if (!enqueue)
1000         talk_shutup(); /* cut off all the pending stuff */
1001
1002     if (n==0)
1003     {   /* special case */
1004         talk_id(VOICE_ZERO, true);
1005         return 0;
1006     }
1007
1008     if (n<0)
1009     {
1010         talk_id(VOICE_MINUS, true);
1011         n = -n;
1012     }
1013
1014     while (n)
1015     {
1016         int segment = n / mil; /* extract in groups of 3 digits */
1017         n -= segment * mil; /* remove the used digits from number */
1018         mil /= 1000; /* digit place for next round */
1019
1020         if (segment)
1021         {
1022             int hundreds = segment / 100;
1023             int ones = segment % 100;
1024
1025             if (hundreds)
1026             {
1027                 talk_id(VOICE_ZERO + hundreds, true);
1028                 talk_id(VOICE_HUNDRED, true);
1029             }
1030
1031             /* combination indexing */
1032             if (ones > 20)
1033             {
1034                int tens = ones/10 + 18;
1035                talk_id(VOICE_ZERO + tens, true);
1036                ones %= 10;
1037             }
1038
1039             /* direct indexing */
1040             if (ones)
1041                 talk_id(VOICE_ZERO + ones, true);
1042
1043             /* add billion, million, thousand */
1044             if (mil)
1045                 talk_id(VOICE_THOUSAND + level, true);
1046         }
1047         level--;
1048     }
1049
1050     return 0;
1051 }
1052
1053 /* Say time duration/interval. Input is time in seconds,
1054    say hours,minutes,seconds. */
1055 static int talk_time_unit(long secs, bool enqueue)
1056 {
1057     int hours, mins;
1058     if (!enqueue)
1059         talk_shutup();
1060     if((hours = secs/3600)) {
1061         secs %= 3600;
1062         talk_value(hours, UNIT_HOUR, true);
1063     }
1064     if((mins = secs/60)) {
1065         secs %= 60;
1066         talk_value(mins, UNIT_MIN, true);
1067     }
1068     if((secs) || (!hours && !mins))
1069         talk_value(secs, UNIT_SEC, true);
1070     else if(!hours && secs)
1071         talk_number(secs, true);
1072     return 0;
1073 }
1074
1075 void talk_fractional(char *tbuf, int value, int unit)
1076 {
1077     int i;
1078     /* strip trailing zeros from the fraction */
1079     for (i = strlen(tbuf) - 1; (i >= 0) && (tbuf[i] == '0'); i--)
1080         tbuf[i] = '\0';
1081
1082     talk_number(value, true);
1083     if (tbuf[0] != 0)
1084     {
1085         talk_id(LANG_POINT, true);
1086         talk_spell(tbuf, true);
1087     }
1088     talk_id(unit, true);
1089 }
1090
1091 int talk_value(long n, int unit, bool enqueue)
1092 {
1093     return talk_value_decimal(n, unit, 0, enqueue);
1094 }
1095
1096 /* singular/plural aware saying of a value */
1097 int talk_value_decimal(long n, int unit, int decimals, bool enqueue)
1098 {
1099     int unit_id;
1100     static const int unit_voiced[] =
1101     {   /* lookup table for the voice ID of the units */
1102         [0 ... UNIT_LAST-1] = -1, /* regular ID, int, signed */
1103         [UNIT_MS]
1104             = VOICE_MILLISECONDS, /* here come the "real" units */
1105         [UNIT_SEC]
1106             = VOICE_SECONDS,
1107         [UNIT_MIN]
1108             = VOICE_MINUTES,
1109         [UNIT_HOUR]
1110             = VOICE_HOURS,
1111         [UNIT_KHZ]
1112             = VOICE_KHZ,
1113         [UNIT_DB]
1114             = VOICE_DB,
1115         [UNIT_PERCENT]
1116             = VOICE_PERCENT,
1117         [UNIT_MAH]
1118             = VOICE_MILLIAMPHOURS,
1119         [UNIT_PIXEL]
1120             = VOICE_PIXEL,
1121         [UNIT_PER_SEC]
1122             = VOICE_PER_SEC,
1123         [UNIT_HERTZ]
1124             = VOICE_HERTZ,
1125         [UNIT_MB]
1126             = LANG_MEGABYTE,
1127         [UNIT_KBIT]
1128             = VOICE_KBIT_PER_SEC,
1129         [UNIT_PM_TICK]
1130             = VOICE_PM_UNITS_PER_TICK,
1131     };
1132
1133     static const int pow10[] = { /* 10^0 - 10^7 */
1134         1, 10, 100, 1000, 10000, 100000, 1000000, 10000000
1135     };
1136
1137     char tbuf[8];
1138     char fmt[] = "%0nd";
1139
1140     if (talk_temp_disable_count > 0)
1141         return -1;  /* talking has been disabled */
1142 #if CONFIG_CODEC != SWCODEC
1143     if (audio_status()) /* busy, buffer in use */
1144         return -1;
1145 #endif
1146
1147     /* special case for time duration */
1148     if (unit == UNIT_TIME)
1149         return talk_time_unit(n, enqueue);
1150
1151     if (unit < 0 || unit >= UNIT_LAST)
1152         unit_id = -1;
1153     else
1154         unit_id = unit_voiced[unit];
1155
1156     if ((n==1 || n==-1) /* singular? */
1157         && unit_id >= VOICE_SECONDS && unit_id <= VOICE_HOURS)
1158     {
1159         unit_id--; /* use the singular for those units which have */
1160     }
1161
1162     /* special case with a "plus" before */
1163     if (n > 0 && (unit == UNIT_SIGNED || unit == UNIT_DB))
1164     {
1165         talk_id(VOICE_PLUS, enqueue);
1166         enqueue = true;
1167     }
1168
1169     if (decimals)
1170     {
1171         /* needed for the "-0.5" corner case */
1172         if (n < 0)
1173         {
1174             talk_id(VOICE_MINUS, enqueue);
1175             n = -n;
1176         }
1177
1178         fmt[2] = '0' + decimals;
1179
1180         snprintf(tbuf, sizeof(tbuf), fmt, n % pow10[decimals]);
1181         talk_fractional(tbuf, n / pow10[decimals], unit_id);
1182
1183         return 0;
1184     }
1185
1186     talk_number(n, enqueue); /* say the number */
1187     talk_id(unit_id, true); /* say the unit, if any */
1188
1189     return 0;
1190 }
1191
1192 /* spell a string */
1193 int talk_spell(const char* spell, bool enqueue)
1194 {
1195     char c; /* currently processed char */
1196
1197     if (talk_temp_disable_count > 0)
1198         return -1;  /* talking has been disabled */
1199 #if CONFIG_CODEC != SWCODEC
1200     if (audio_status()) /* busy, buffer in use */
1201         return -1;
1202 #endif
1203
1204     if (!enqueue)
1205         talk_shutup(); /* cut off all the pending stuff */
1206
1207     while ((c = *spell++) != '\0')
1208     {
1209         /* if this grows into too many cases, I should use a table */
1210         if (c >= 'A' && c <= 'Z')
1211             talk_id(VOICE_CHAR_A + c - 'A', true);
1212         else if (c >= 'a' && c <= 'z')
1213             talk_id(VOICE_CHAR_A + c - 'a', true);
1214         else if (c >= '0' && c <= '9')
1215             talk_id(VOICE_ZERO + c - '0', true);
1216         else if (c == '-')
1217             talk_id(VOICE_MINUS, true);
1218         else if (c == '+')
1219             talk_id(VOICE_PLUS, true);
1220         else if (c == '.')
1221             talk_id(VOICE_DOT, true);
1222         else if (c == ' ')
1223             talk_id(VOICE_PAUSE, true);
1224         else if (c == '/')
1225             talk_id(VOICE_CHAR_SLASH, true);
1226     }
1227
1228     return 0;
1229 }
1230
1231 void talk_disable(bool disable)
1232 {
1233     if (disable)
1234         talk_temp_disable_count++;
1235     else
1236         talk_temp_disable_count--;
1237 }
1238
1239 void talk_setting(const void *global_settings_variable)
1240 {
1241     const struct settings_list *setting;
1242     if (!global_settings.talk_menu)
1243         return;
1244     setting = find_setting(global_settings_variable, NULL);
1245     if (setting == NULL)
1246         return;
1247     if (setting->lang_id)
1248         talk_id(setting->lang_id,false);
1249 }
1250
1251
1252 #if CONFIG_RTC
1253 void talk_date(const struct tm *tm, bool enqueue)
1254 {
1255     talk_id(LANG_MONTH_JANUARY + tm->tm_mon, enqueue);
1256     talk_number(tm->tm_mday, true);
1257     talk_number(1900 + tm->tm_year, true);
1258 }
1259
1260 void talk_time(const struct tm *tm, bool enqueue)
1261 {
1262     if (global_settings.timeformat == 1)
1263     {
1264         /* Voice the hour */
1265         long am_pm_id = VOICE_AM;
1266         int hour = tm->tm_hour;
1267         if (hour >= 12)
1268         {
1269             am_pm_id = VOICE_PM;
1270             hour -= 12;
1271         }
1272         if (hour == 0)
1273             hour = 12;
1274         talk_number(hour, enqueue);
1275
1276         /* Voice the minutes */
1277         if (tm->tm_min == 0)
1278         {
1279             /* Say o'clock if the minute is 0. */
1280             talk_id(VOICE_OCLOCK, true);
1281         }
1282         else
1283         {
1284             /* Pronounce the leading 0 */
1285             if(tm->tm_min < 10)
1286                 talk_id(VOICE_OH, true);
1287             talk_number(tm->tm_min, true);
1288         }
1289         talk_id(am_pm_id, true);
1290     }
1291     else
1292     {
1293         /* Voice the time in 24 hour format */
1294         talk_number(tm->tm_hour, enqueue);
1295         if (tm->tm_min == 0)
1296         {
1297             talk_id(VOICE_HUNDRED, true);
1298             talk_id(VOICE_HOUR, true);
1299         }
1300         else
1301         {
1302             /* Pronounce the leading 0 */
1303             if(tm->tm_min < 10)
1304                 talk_id(VOICE_OH, true);
1305             talk_number(tm->tm_min, true);
1306         }
1307     }
1308 }
1309
1310 #endif /* CONFIG_RTC */