Top-level README file for the JEIDA (JCSD) Japanese Speech Corpus ----------------------------------------------------------------- The JEIDA corpus consists of high-fidelity recordings of 150 native speakers of Japanese; each speaker produces four repetitions of 323 short prompts, including city names, control words, monosyllabic words, isolated digits, and strings of four digits. Each reading session was recorded with two microphones, yielding two channels that differ in audio quality for each utterance. The total number of speech data files intended for the corpus is therefore: 150 * 323 * 4 * 2 = 387,600 speakers prompts repetitions channels speech files The actual number of speech files in the corpus is somewhat less: there were gaps in the data collection that eliminated nearly 13,000 files; in particular, for 10 of the speakers, only one channel was recorded; for a handful of other speakers, a few of the utterances are missing on both channels because of anomalies during the recording sessions. Refer to the main documentation and tables for details. For purposes of publication by the LDC, the corpus has been organized onto 44 CD-ROMs; the partitioning of the data files has been done primarily by channel (22 CD-ROMs each for channel 0 and channel 1), and secondarily by category of prompts. The volume name of each disc reflects channel and prompt category. The prompt category is also reflected in the name of the top-level data directory on each disc. The main data directories contain subdirectories for each speaker; all speaker directories have three-character names, consisting of the speaker sex (``m'' or ``f'') and a two-digit identifier (female speakers are numbered f01 through f75, while male speakers are numbered m01 through m77, but with m26 and m30 not present in the corpus). Within each speaker directory are the individual speech files; the file names reflect the speaker, the prompt category and prompt index number, the repetition number, and the channel, as follows: SSSPNNNR.spC where: SSS = speaker id (e.g. "f01") P = prompt type (one of "a,b,c,d,i,m,n") NNN = prompt number (starting at "001") R = repetition number (one of "a,b,c,d") C = channel number ("0" or "1") Example: f01a001d.sp0 means: Speaker f01 Control Words A Prompt #001 Fourth repetition Channel 0 The table below lists the volume names, the top-level data directory names (plus single-character prompt letter in the file names), and the range of speakers for each disc in the set; in the volume name field of the table, the ``?'' character represents the channel number for the disc (either ``0'' or ``1''). Main Data Approx. Volume Name Directory Speakers Size --------------------------------------------------------------- JEIDA?_CNTRLA_1 cntrl_a (a) f01-m77 420 MB JEIDA?_CNTRLB_1 cntrl_b (b) f01-f75 430 MB JEIDA?_CNTRLB_2 cntrl_b (b) m01-m77 415 MB JEIDA?_CNTRLC_1 cntrl_b (c) f01-f75 405 MB JEIDA?_CNTRLC_2 cntrl_b (c) m01-m77 395 MB JEIDA?_FOURDGT_1 four_dgt (d) f01-f50 545 MB JEIDA?_FOURDGT_2 four_dgt (d) m01-m52 535 MB JEIDA?_FOURDGT_3 four_dgt (d) f51-f75,m53-m77 545 MB JEIDA?_ISOLDGT_1 isol_dgt (i) f01-m77 415 MB JEIDA?_MONOSYL_1 mono_syl (m) f01-f25 475 MB JEIDA?_MONOSYL_2 mono_syl (m) f26-f50 490 MB JEIDA?_MONOSYL_3 mono_syl (m) f51-f75 515 MB JEIDA?_MONOSYL_4 mono_syl (m) m01-m25 480 MB JEIDA?_MONOSYL_5 mono_syl (m) m27-m52 495 MB JEIDA?_MONOSYL_6 mono_syl (m) m53-m77 485 MB JEIDA?_CITY_1 city (n) f01-f25 550 MB JEIDA?_CITY_2 city (n) f26-f50 550 MB JEIDA?_CITY_3 city (n) f51-f75 580 MB JEIDA?_CITY_4 city (n) m01-m25 545 MB JEIDA?_CITY_5 city (n) m27-m52 560 MB JEIDA?_CITY_6 city (n) m53-m77 540 MB --------------------------------------------------------------- Note that the file name example given above, "f01a001d.sp0", will be found on volume "JEIDA0_CNTRLA_1" in the directory path "cntrl_a/f01". Each CD-ROM contains a "document" directory, whose content is identical on all volumes; this directory contains the main documentation for the corpus (in the form of a PostScript file), together with a set of ASCII-formatted tables extracted from the main documentation.