Răsfoiți Sursa

fix m2met2 baseline bugs

yhliang 2 ani în urmă
părinte
comite
db77a41e29

+ 4 - 4
egs/alimeeting/sa-asr/asr_local.sh

@@ -1153,10 +1153,10 @@ if ! "${skip_train}"; then
         mkdir -p ${sa_asr_exp}/log
         INIT_FILE=${sa_asr_exp}/ddp_init
         
-        if [ ! -f "exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pth" ]; then
+        if [ ! -f "exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pb" ]; then
             # download xvector extractor model file
             python local/download_xvector_model.py exp
-            log "Successfully download the pretrained xvector extractor to exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pth"
+            log "Successfully download the pretrained xvector extractor to exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pb"
         fi
         
         if [ -f $INIT_FILE ];then
@@ -1195,8 +1195,8 @@ if ! "${skip_train}"; then
                     --init_param "${asr_exp}/valid.acc.ave.pb:decoder.decoders.3:decoder.decoder4.2" \
                     --init_param "${asr_exp}/valid.acc.ave.pb:decoder.decoders.4:decoder.decoder4.3" \
                     --init_param "${asr_exp}/valid.acc.ave.pb:decoder.decoders.5:decoder.decoder4.4" \
-                    --init_param "exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pth:encoder:spk_encoder"   \
-                    --init_param "exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pth:decoder:spk_encoder:decoder.output_dense"   \
+                    --init_param "exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pb:encoder:spk_encoder"   \
+                    --init_param "exp/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/sv.pb:decoder:spk_encoder:decoder.output_dense"   \
                     --valid_data_path_and_name_and_type "${_asr_valid_dir}/${_scp},speech,${_type}" \
                     --valid_data_path_and_name_and_type "${_asr_valid_dir}/text,text,text" \
                     --valid_data_path_and_name_and_type "${_asr_valid_dir}/oracle_profile_nopadding.scp,profile,npy" \

+ 6 - 5
egs/alimeeting/sa-asr/local/alimeeting_data_prep.sh

@@ -61,9 +61,9 @@ mkdir -p $near_dir
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then 
     log "stage 1:process alimeeting near dir"
     
-    find -L $near_raw_dir/audio_dir -iname "*.wav" >  $near_dir/wavlist
+    find -L $near_raw_dir/audio_dir -iname "*.wav" | sort >  $near_dir/wavlist
     awk -F '/' '{print $NF}' $near_dir/wavlist | awk -F '.' '{print $1}' > $near_dir/uttid   
-    find -L $near_raw_dir/textgrid_dir  -iname "*.TextGrid" > $near_dir/textgrid.flist
+    find -L $near_raw_dir/textgrid_dir  -iname "*.TextGrid" | sort > $near_dir/textgrid.flist
     n1_wav=$(wc -l < $near_dir/wavlist)
     n2_text=$(wc -l < $near_dir/textgrid.flist)
     log  near file found $n1_wav wav and $n2_text text.
@@ -90,9 +90,9 @@ fi
 if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     log "stage 2:process alimeeting far dir"
     
-    find -L $far_raw_dir/audio_dir -iname "*.wav" >  $far_dir/wavlist
+    find -L $far_raw_dir/audio_dir -iname "*.wav" | sort >  $far_dir/wavlist
     awk -F '/' '{print $NF}' $far_dir/wavlist | awk -F '.' '{print $1}' > $far_dir/uttid   
-    find -L $far_raw_dir/textgrid_dir  -iname "*.TextGrid" > $far_dir/textgrid.flist
+    find -L $far_raw_dir/textgrid_dir  -iname "*.TextGrid" | sort > $far_dir/textgrid.flist
     n1_wav=$(wc -l < $far_dir/wavlist)
     n2_text=$(wc -l < $far_dir/textgrid.flist)
     log  far file found $n1_wav wav and $n2_text text.
@@ -120,7 +120,8 @@ fi
 
 if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
     log "stage 3: finali data process"
-
+    local/fix_data_dir.sh $near_dir
+    local/fix_data_dir.sh $far_dir
     local/copy_data_dir.sh $near_dir data/${tgt}_Ali_near
     local/copy_data_dir.sh $far_dir data/${tgt}_Ali_far
 

+ 1 - 1
egs/alimeeting/sa-asr/path.sh

@@ -2,4 +2,4 @@ export FUNASR_DIR=$PWD/../../..
 
 # NOTE(kan-bayashi): Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
 export PYTHONIOENCODING=UTF-8
-export PATH=$FUNASR_DIR/funasr/bin:$PATH
+export PATH=$FUNASR_DIR/funasr/bin:./utils:$PATH