Dear Adam,
When I use qnstrn to get TRAP feature, it prints
"(QN_InFtrStream_RandWindow.ftr1_file): WARNING - Segment 2292 has
only 90 frames, which is too small - no windows will be produced from
this segment."

Now I use 101 frames to calculate TRAP feature. Is there anything
wrong with my configure? This is my configure at below:
${qnstrn:=/public/DISK3/QQZHANG/xcyao/nn/quicknet-v3_11/qnstrn} \
ftr1_file=${ftr1_file:=$pfile} \
ftr1_format=${ftr1_format:="pfile"} \
ftr1_width=${ftr1_width:=0} \
ftr2_file=${ftr2_file:=""} \
ftr2_format=${ftr2_format:="pfile"} \
ftr2_width=${ftr2_width:=0} \
unary_file=${unary_file:=""} \
ftr1_norm_file=${ftr1_norm_file:=$normfile} \
ftr2_norm_file=${ftr2_norm_file:=""} \
ftr1_ftr_start=${ftr1_ftr_start:=14} \
ftr1_ftr_count=${ftr1_ftr_count:=1} \
ftr2_ftr_start=${ftr2_ftr_start:=0} \
ftr2_ftr_count=${ftr2_ftr_count:=0} \
ftr1_delta_order=${ftr1_delta_order:=0} \
ftr1_delta_win=${ftr1_delta_win:=101} \
ftr1_norm_mode=${ftr1_norm_mode:=online} \
ftr1_norm_alpha_m=${ftr1_norm_alpha_m:=0.005} \
ftr1_norm_alpha_v=${ftr1_norm_alpha_v:=0.005} \
ftr2_delta_order=${ftr2_delta_order:=0} \
ftr2_delta_win=${ftr2_delta_win:=101} \
ftr2_norm_mode=${ftr2_norm_mode:="file"} \
ftr2_norm_alpha_m=${ftr2_norm_alpha_m:=0.005} \
ftr2_norm_alpha_v=${ftr2_norm_alpha_v:=0.005} \
ftr1_window_offset=${ftr1_window_offset:=0} \
ftr1_window_len=${ftr1_window_len:=101} \
ftr2_window_offset=${ftr2_window_offset:=0} \
ftr2_window_len=${ftr2_window_len:=101} \
unary_window_offset=${unary_window_offset:=3} \
hardtarget_file=${hardtarget_file:=$ilabfile} \
hardtarget_format=${hardtarget_format:="ilab"} \
hardtarget_window_offset=${hardtarget_window_offset:=50} \
hardtarget_lastlab_reject=${hardtarget_lastlab_reject:=0} \
softtarget_file=${softtarget_file:=""} \
softtarget_format=${softtarget_format:=""} \
softtarget_width=${softtarget_width:=0} \
softtarget_window_offset=${softtarget_window_offset:=50} \
window_extent=${window_extent:=101} \
train_cache_frames=${train_cache_frames:=10000000} \
train_cache_seed=${train_cache_seed:=0} \
$TRAIN_RANGE_SPEC \
$CV_RANGE_SPEC \
init_weight_file=${init_weight_file:=""} \
out_weight_file=${out_weight_file:=cts.train.15_TRAP.qnsout.weights} \
log_weight_file=${log_weight_file:=cts.train.15_TRAP.qnslog.weights} \
init_random_bias_min=${init_random_bias_min:=-4.1} \
init_random_bias_max=${init_random_bias_max:=-3.9} \
init_random_weight_min=${init_random_weight_min:=-0.1} \
init_random_weight_max=${init_random_weight_max:=0.1} \
init_random_seed=${init_random_seed:=0} \
learnrate_schedule=${learnrate_schedule:=newbob} \
learnrate_vals=${learnrate_vals:=0.001} \
learnrate_epochs=${learnrate_epochs:=9999} \
learnrate_scale=${learnrate_scale:=0.5} \
unary_size=${unary_size:=0} \
mlp3_input_size=${mlp3_input_size:=101} \
mlp3_hidden_size=${mlp3_hidden_size:=500} \
mlp3_output_size=${mlp3_output_size:=49} \
mlp3_output_type=${mlp3_output_type:=softmax} \
mlp3_pp=${mlp3_pp:=true} \
mlp3_blas=${mlp3_blas:=true} \
mlp3_bunch_size=${mlp3_bunch_size:=16} \
mlp3_threads=${mlp3_threads:=2} \
log_file=${log_file:="log_15.train.txt"} \
verbose=${verbose:=true} \
debug=${debug:=1}

Actually, when I use 9 frame PLP to calculate PLP_MLP feature,does the
PLP_MLP feature have the same frame number with the input PLP feature,
or 8 frames less than PLP feature?
If I want to get the PLP_MLP feature which has the same frame number
with the input PLP feature, how can I set the configures?

Thank you very much!