man.diff

speech signal process tools

or the corresponding parameter-file entries.
This option overrides any value for
.I spec_type
specified in the parameter file.
.TP
.BI \-W " channel_width" "\fR [(see text)]\fP"
The total bandwidth in mel of each triangular filter\*-i.e.
the width of the base of the triangle.
If the number of filters is specified explicitly (option
.BR \-n ,
parameter
.IR num_freqs )
then the default width is chosen so that the given number of bands
will cover the specified range (see
.BR \-M )
with 50% overlap between adjacent bands.
In that case the width is
.RI 2( mel_high " \- " mel_low\c
.RI )/( num_freqs " + 1)."
A specified value of 0 for
.I channel_width
implies use of this default value.
If the number of filters is not specified
then the channel bandwidth must be given explicitly,
either with this option or by the parameter
.I channel_width
in the parameter file.
This option overrides the parameter-file value.
.TP
.B \-X
This option causes the program to write a frequency table
on the standard error output.
The table lists
the filter peak frequencies and band edges in mels and in hertz.
This output is also included in the debug output produced by option
.B \-x
with a
.I debug_level
of 2 or higher.
.SH ESPS PARAMETERS
.PP
The following parameters may be read, if present, from the parameter file.
.TP
.I "add_const \- float"
.IP
Constant to be added to mel-spectral powers
before they are written out.
See option
.BR \-a .
(See also option
.B \-m
and parameter
.IR mult_const. )
This parameter is not read if
.B \-a
is specified.
.TP
.I "band_high \- float"
.IP
High limit of the range of frequencies (in hertz) to be covered.
See option
.BR \-H .
(See also option
.B \-M
and parameters
.IR band_low,
.IR mel_high,
and
.IR mel_low. )
This parameter is not read if option
.B \-M
or
.B \-H
is specified or if parameter
.I mel_high
is also specified.
A value of 0 implies use of the default value\*-see
.BR \-H .
.TP
.I "band_low \- float"
.IP
Low limit of the range of frequencies (in hertz) to be covered.
See option
.BR \-H .
(See also option
.B \-M
and parameters
.IR band_high,
.IR mel_high,
and
.IR mel_low. )
This parameter is not read if option
.B \-M
or
.B \-H
is specified or if parameter
.I mel_low
is also specified.
.TP
.I "mel_high \- float"
.IP
High limit of the range of mel-scale values to be covered.
See option
.BR \-M .
(See also option
.B \-H
and parameters
.IR mel_low,
.IR band_high,
and
.IR band_low. )
This parameter is not read if
.B \-M
or
.B \-H
is specified.
It takes precedence if parameter
.I band_high
is also specified.
A value of 0 implies use of the default value\*-see
.BR \-M .
.TP
.I "mel_low \- float"
.IP
Low limit of the range of mel-scale values to be covered.
See option
.BR \-M .
(See also option
.B \-H
and parameters
.IR mel_high,
.IR band_low,
and
.IR band_high. )
This parameter is not read if option
.B \-M
or
.B \-H
is specified.
It takes precedence if parameter
.I band_low
is also specified.
.TP
.I "mult_const \- float"
.IP
Constant by which mel-spectral powers are to be multiplied
before being written out.
See option
.BR \-m .
(See also option
.BR \-a
and parameter
.IR add_const. )
This parameter is not read if
.B \-m
is specified.
.TP
.I "nan \- int"
.IP
The number of records to process.
See option
.BR \-r .
(See also parameter
.IR start. )
A value of 0 implies processing all records from
.I start
to the end of the file; this is the default.
This parameter is not read if
.B \-r
is specified.
.TP
.I "num_freqs \- int"
.IP
The number of mel values at which spectral values will be computed.
See option
.BR \-n .
A value of 0 implies use of the default value (see
.BR \-n ).
This parameter is not read if the option is specified.
.TP
.I "spec_type \- string"
.IP
Allowed values are "DB" and "PWR" (case-insensitive).
These indicate whether to output mel-spectral powers
directly ("PWR") or convert them to dB ("DB").
See option
.BR \-S .
This parameter is not read if the option is specified.
.TP
.I "start \- int"
.IP
The number of the first input record to process.
See option
.BR \-r .
(See also parameter
.IR nan. )
This parameter is not read if
.B \-r
is specified.
.TP
.I "channel_width \- float"
.IP
The total bandwidth in mel of each triangular filter.
See option
.BR \-W .
This parameter is not read if the option is specified.
A value of 0 implies use of the default value (see
.BR \-W ).
.SH ESPS COMMON
.PP
The ESPS common file is not accessed.
.SH ESPS HEADERS
.PP
The output file header is an ESPS FEA_SPEC header (see
.IR FEA_SPEC (5\-ESPS),
.IR init_feaspec_hd (3\-ESPS)).
A copy of the input header is included in the output file
as a source header.
The output file is tagged if and only if the input file is tagged.
If the files are tagged, the
.I refer
file name in the output header is copied from the input header,
and the generic header item
.I src_sf
is copied from the input header if present;
otherwise the value of
.I sf
is used.
Output generic header items
.I start_time
and
.I record_freq
are determined by the corresponding input header items, if present,
and by the
.I start
record number.
.PP
Generic header items
.I add_const,
.I band_high,
.I band_low,
.I channel_width,
.I mel_high,
.I mel_low,
.I mult_const,
.I nan,
and
.I start
are added to the output file header to record the values
specified with the corresponding parameters and options.
The data types of the header items are DOUBLE for
.I float
parameters and LONG for
.I int
parameters.
The usual FEA_SPEC generic header item
.I spec_type
(see
.IR FEA_SPEC (5\-ESPS))
takes the CODED value SPTYP_DB or SPTYP_PWR,
as determined by the parameter
.I spec_type
or the argument of
.BR \-S ,
and the usual FEA_SPEC generic header item
.I num_freqs
takes the value of
.I num_freqs.
.PP
A special generic header item
.I mel_freqs
of type FLOAT
is added to record the selected set of uniformly spaced mel values
.I m
for which mel-spectral values are computed.
The equivalent linear frequencies (nonuniformly spaced)
are recorded in the usual FEA_SPEC generic header item
.I freqs
(see
.IR FEA_SPEC (5\-ESPS)).
.SH FUTURE CHANGES
.PP
None contemplated.
.SH EXAMPLES
.PP
.SH ERRORS AND DIAGNOSTICS
.PP
The program prints a synopsis of command-line usage and exits
if an unknown option is specified
or if the number of file names is wrong.
It exits with an error message
if
.I input.spec
and
.I output.spec
are the same (but not ``\-'');
if
.I input.spec
does not exist or is not an ESPS FEA_SPEC file;
if
.I freq_format
in the input file is not SYM_EDGE;
if the range of record numbers (see
.BR \-r )
is empty or starts before the beginning of the file;
if the mel-scale range to be covered (see
.B \-H
and
.BR \-M )
is less than
.I channel_width
(see
.BR \-W );
if
.B \-H
and
.B \-M
are both specified;
if neither
.I channel_width
nor
.I num_freqs
(see
.BR \-n )
is specified;
if
.I channel_width
is 0 or negative;
if
.I num_freqs
(see
.BR \-n )
is less than 1;
if
.I num_freqs
is 2 or more, but the uniform interval between mel values
(see
.BR \-M )
is 0 or less;
if any of these mel values corresponds to a frequency
less than 0 or greater than the Nyquist frequency;
or if an unsupported
.I spec_type
(see
.BR \-S )
is specified.
.SH BUGS
.PP
None known.
.SH REFERENCES
.PP
[1] Steve Young, Julian Odell, Dave Ollason, Valtcho Valtchev,
and Phil Woodland,
.I "The HTK Book,"
Entropic, 1997.
.SH "SEE ALSO"
.IR barkspec (1\-ESPS),
.IR fft (1\-ESPS),
.IR me_spec (1\-ESPS),
.br
.IR init_feaspec_hd (3\-ESPS),
.IR FEA_SPEC (5\-ESPS)
.SH AUTHOR
Rodney Johnson