The NIST LRE22 Dev Data Package, originally released to evaluation participants as:
LDC2022E14 2022 NIST Language Recognition Evaluation Development Data,
consists of:

1. The data in, "./data/dev", which contains 4200 audio files, each sampled at
8khz and encoded as 8-bit a-law in NIST SPHERE format (.sph).  There are 300
files for each of the 14 LRE22 target languages.  Audio files contain
nominally 3-30 seconds of speech (as determined by an automatic speech
activity detector) with duration ranging from ~3s to ~93s, with a median
duration of ~24s.  Note that multiple audio files may have been derived from a
common source audio file (this can be determined using the provided metadata).

2. In "./docs", the LRE22 evaluation plan; a list of the MD5 checksums and
names of the 4200 dev data files; and a tab-separated file,
"./metadata/lre22_dev_metadata.tsv". This metadata file contains the following
information for each audio file:

     a. "file_name": The SPH file name in the "./data/dev" folder.
     b. "language_code": Language code for the language spoken in the audio
         file. This is a value from the list of 14 target languages in LRE22.
     c. "ldc_audioid": This is the LDC-provided audio id for the file which
           was used to generate this SPH file. Multiple files may have been
           derived from a single source audio file, in which case they share
           the same ldc_audioid.
     d. "time_offset_sec": This is the time offset in seconds in the
           ldc_audioid file from which the current file was generated.
     e. "duration_sec": This is the duration in seconds of the current file.

4. This file, "./README_dev.txt"