sphfile

Numpy-based NIST SPH audio-file reader. This is for use with NIST SPH audio-files, the most likely use being extracting the TEDLIUM_release2 audio into formats that standard tools can easily process.

Note that this library doesn't require any external tools such as vox or gstreamer. It just loads the data into a numpy array and then lets you dump it back out to wave files.

Note that the library does not support files with embedded-shorten-* encodings, only the base ulaw encoding. You will need to convert such files with:

sph2pipe file.sph file-raw.sph

to allow them to be loaded.

Usage

from sphfile import SPHFile
sph =SPHFile(
    'TEDLIUM_release2/test/sph/JamesCameron_2010.sph'
)
# Note that the following loads the whole file into ram
print( sph.format )
# write out a wav file with content from 111.29 to 123.57 seconds
sph.write_wav( 'test.wav', 111.29, 123.57 )

Requirements

• numpy

License

MIT License (c) 2017 Mike C. Fletcher

Authors

sphfile was written by Mike C. Fletcher.

History

• 1.0.3 -- Allow for other header keys during header format parsing
• 1.0.2 -- Use signed integers for 2 and 4-byte sample_n_bytes
• 1.0.1 -- Fix to allow for files that have non-sample-multiple bytes in the data section
• 1.0.0 -- Initial release