The FeatureSet Object - Tcl methods
The FeatureSet Object
Tcl methods
Basic methods
SVector ,
FMatrix ,
delete
- Define and delete Features.
puts ,
display
Show info and content.
JANUS interface methods
setDesc,
setAccess,
eval
- Define 'Feature Description', 'Access Description' and
do preprocessing.
ADC methods (SVector features)
readADC ,
writeADC
- Read/Write Audio files (SVector features).
adc2pow ,
peak ,
maxpeak ,
zero
- Frame based analysis of power, peak values and zero crossing rate.
silTK
- Speech/silence detector.
spectrum,
adc2mel
- Frame based spectrum and melscale analysis.
tone,
noise
- Create signal and noise samples.
xtalk
- Use adaptive filter to remove xtalk.
downsample
- Downsample 16bit,16kHz to telephone quality 8bit,8kHz.
Spectrum, filterbank methods
melscale,
auditory,
postaud,
plp
- Calculating filterbank and cepstral coefficients from spectrum
Feature I/O
read,
write
- Read/Write FMatrix features.
Methods using one source
split,
cut
- Take some coefficients and cut out some frames (or samples).
flip,
delta,
shift,
adjacent
- Flip feature. Calculate delta coefficients. Shift frames. Create
supervector with adjacent frames.
lin,
log,
exp,
pow
- Simple functions of the feature coefficients.
filter,
normframe,
meansub,
offset,
normalize,
thresh
- Filter features. Normalize frames. Do Meansubtraction, adaptive offset removal. Normalize
feature channels. Cut off values exceeding a threshold.
cdcn
- Codebook-dependent Cepstral Normalisation
Methods using more than one source
merge,
concat
- Merging and concatenating features together.
add,
mul,
matmul
- Add, multiply 2 features and multiply feature with matrix.
maier@ira.uka.de