Merge pull request espnet#3810 from espnet/ftshijt-patch-1

Some minor fixes for commonvoice recipe
shinichiye · Nov 23, 2021 · 0d8cd47 · 0d8cd47
2 parents 288a5c1 + a20138b
commit 0d8cd47
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 5 deletions.
diff --git a/egs/commonvoice/asr1/local/data_prep.pl b/egs/commonvoice/asr1/local/data_prep.pl
@@ -68,6 +68,6 @@
   die "Error creating spk2utt file in directory $out_dir";
 }
 system("env LC_COLLATE=C utils/fix_data_dir.sh $out_dir");
-if (system("env LC_COLLATE=C utils/validate_data_dir.sh --no-feats $out_dir") != 0) {
+if (system("env LC_COLLATE=C utils/validate_data_dir.sh --non-print --no-feats $out_dir") != 0) {
   die "Error validating directory $out_dir";
 }
diff --git a/egs2/commonvoice/asr1/conf/tuning/train_asr_transformer.yaml b/egs2/commonvoice/asr1/conf/tuning/train_asr_transformer.yaml
@@ -48,4 +48,5 @@ best_model_criterion:
    - max
 keep_nbest_models: 10
 
-init: chainer
+init: xavier_uniform # Our empirical studies shows that this initialization
+                     # is very important to low-resource ASR training
diff --git a/egs2/commonvoice/asr1/local/data.sh b/egs2/commonvoice/asr1/local/data.sh
@@ -24,7 +24,7 @@ log() {
     echo -e "$(date '+%Y-%m-%dT%H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
 }
 
-mkdir ${COMMONVOICE}
+mkdir -p ${COMMONVOICE}
 if [ -z "${COMMONVOICE}" ]; then
     log "Fill the value of 'COMMONVOICE' of db.sh"
     exit 1
@@ -44,7 +44,8 @@ log "data preparation started"
 
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then 
     log "stage1: Download data to ${COMMONVOICE}"
-    mkdir -p ${COMMONVOICE}
+    log "The default data of this recipe is from commonvoice 5.1, for newer version, you need to register at \
+         https://commonvoice.mozilla.org/"
     local/download_and_untar.sh ${COMMONVOICE} ${data_url} ${lang}.tar.gz
 fi
 

diff --git a/tools/Makefile b/tools/Makefile
@@ -57,6 +57,9 @@ endif
 
 #########################################
 
+bc.done: activate_python.sh
+	. ./activate_python.sh && { command -v bc || conda install -y bc -c conda-forge; }
+	touch bc.done
 cmake.done: activate_python.sh
 	. ./activate_python.sh && { command -v cmake || conda install -y cmake; }
 	touch cmake.done
@@ -73,7 +76,7 @@ sndfile.done: activate_python.sh
 	. ./activate_python.sh && { python3 -c "from ctypes.util import find_library as F; assert F('sndfile') is not None" || conda install -y libsndfile=1.0.28 -c conda-forge; }
 	touch sndfile.done
 ifneq ($(strip $(USE_CONDA)),)
-conda_packages.done: cmake.done flac.done ffmpeg.done sox.done sndfile.done
+conda_packages.done: bc.done cmake.done flac.done ffmpeg.done sox.done sndfile.done
 else
 conda_packages.done:
 endif