sphinx_fe - Man Page

Convert audio files to acoustic feature files


sphinx_fe [ options ]...


This program converts audio files (in either Microsoft WAV, NIST Sphere, or raw format) to acoustic feature files for input to batch-mode speech recognition.  The resulting files are also useful for various other things.  A list of options follows:


Preemphasis parameter


file (e.g. feat.params from an acoustic model) to read parameters from.  This will override anything set in other command line arguments.


Number of samples to read at a time.


Create missing subdirectories in output directory


file for batch processing


Input is cepstral files, output is log spectral files


directory, input file names are relative to this, if defined


Add 1/2-bit noise


directory, output files are relative to this


Use double bandwidth filters (same center freq)


extension to be applied to all input files


extension to be applied to all output files


Shows example of how to use the tool


Frame rate


Shows the usage of the tool


audio input file


Endianness of input data, big or little, ignored if NIST or MS Wav


Length of sin-curve for liftering, or 0 for no liftering.


Write out logspectral files instead of cepstra


Lower edge of filters


Endianness of machine, big or little


Defines input format as Microsoft Wav (RIFF)


Number of cep coefficients


Number of channels of data (interlaced samples assumed)


Size of FFT


Number of filter banks


Defines input format as NIST sphere


Number of parts to run in (supersedes -nskip and -runlen if non-zero)


If a control file was specified, the number of utterances to skip at the head of the file


cepstral output file


Format of output files - one of sphinx, htk, text.


Index of the part to run (supersedes -nskip and -runlen if non-zero)


Defines input format as raw binary data


Remove DC offset from each frame


Remove noise with spectral subtraction in mel-energies


Enables VAD, removes silence frames from processing


Round mel filter frequencies to DFT points


If a control file was specified, the number of utterances to process, or -1 for all


Sampling rate


Seed for random number generator; if less than zero, pick our own


Write out cepstral-smoothed logspectral files


Input is log spectral files, output is cepstral files


Input is NIST sphere (possibly with Shorten), use sph2pipe to convert


Which type of transform to use to calculate cepstra (legacy, dct, or htk)


Normalize mel filters to unit area


Upper edge of filters


Num of silence frames to keep after from speech to silence.


Num of speech frames to keep before silence to speech.


Num of speech frames to trigger vad from silence to speech.


Threshold for decision between noise and silence frames. Log-ratio between signal level and noise level.


Show input filenames


defining the warping function


Warping function type (or shape)


Channel to process (numbered from 1), or 0 to mix all channels


Hamming window length

Currently the only kind of features supported are MFCCs (mel-frequency cepstral coefficients).  There are numerous options which control the properties of the output features.  It is VERY important that you document the specific set of flags used to create any given set of feature files, since this information is NOT recorded in the files themselves, and any mismatch between the parameters used to extract features for recognition and those used to extract features for training will cause recognition to fail.


Written by numerous people at CMU from 1994 onwards.  This manual page by David Huggins-Daines <dhuggins@cs.cmu.edu>

Referenced By

pocketsphinx(1), pocketsphinx_batch(1).