update repo

egs/aishell/paraformerbert/local/extract_embeds.sh

@@ -5,8 +5,6 @@ stop_stage=3
 
 bert_model_root="../../huggingface_models"
 bert_model_name="bert-base-chinese"
-#bert_model_name="chinese-roberta-wwm-ext"
-#bert_model_name="mengzi-bert-base"
 raw_dataset_path="../DATA"
 model_path=${bert_model_root}/${bert_model_name}
 
@@ -16,7 +14,7 @@ nj=32
 
 for data_set in train dev test;do
     scp=$raw_dataset_path/dump/fbank/${data_set}/text
-    local_scp_dir_raw=$raw_dataset_path/embeds/$bert_model_name/${data_set}
+    local_scp_dir_raw=${raw_dataset_path}/${data_set}
     local_scp_dir=$local_scp_dir_raw/split$nj
     local_records_dir=$local_scp_dir_raw/ark
 

egs/aishell/paraformerbert/run.sh

@@ -16,12 +16,11 @@ infer_cmd=utils/run.pl
 feats_dir="../DATA" #feature output dictionary, for large data
 exp_dir="."
 lang=zh
-dumpdir=dump/fbank
-feats_type=fbank
 token_type=char
-scp=feats.scp
-type=kaldi_ark
-stage=0
+type=sound
+scp=wav.scp
+speed_perturb="0.9 1.0 1.1"
+stage=3
 stop_stage=4
 
 skip_extract_embed=false
@@ -30,15 +29,14 @@ bert_model_name="bert-base-chinese"
 
 # feature configuration
 feats_dim=80
-sample_frequency=16000
-nj=32
-speed_perturb="0.9,1.0,1.1"
+nj=64
 
 # data
-data_aishell=
+raw_data=
+data_url=www.openslr.org/resources/33
 
 # exp tag
-tag=""
+tag="exp1"
 
 . utils/parse_options.sh || exit 1;
 
@@ -53,7 +51,7 @@ valid_set=dev
 test_sets="dev test"
 
 asr_config=conf/train_asr_paraformerbert_conformer_12e_6d_2048_256.yaml
-model_dir="baseline_$(basename "${asr_config}" .yaml)_${feats_type}_${lang}_${token_type}_${tag}"
+model_dir="baseline_$(basename "${asr_config}" .yaml)_${lang}_${token_type}_${tag}"
 
 inference_config=conf/decode_asr_transformer_noctc_1best.yaml
 inference_asr_model=valid.acc.ave_10best.pb
@@ -70,10 +68,17 @@ else
     _ngpu=0
 fi
 
+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
+    echo "stage -1: Data Download"
+    local/download_and_untar.sh ${raw_data} ${data_url} data_aishell
+    local/download_and_untar.sh ${raw_data} ${data_url} resource_aishell
+fi
+
+
 if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     echo "stage 0: Data preparation"
     # Data preparation
-    local/aishell_data_prep.sh ${data_aishell}/data_aishell/wav ${data_aishell}/data_aishell/transcript ${feats_dir}
+    local/aishell_data_prep.sh ${raw_data}/data_aishell/wav ${raw_data}/data_aishell/transcript ${feats_dir}
     for x in train dev test; do
         cp ${feats_dir}/data/${x}/text ${feats_dir}/data/${x}/text.org
         paste -d " " <(cut -f 1 -d" " ${feats_dir}/data/${x}/text.org) <(cut -f 2- -d" " ${feats_dir}/data/${x}/text.org | tr -d " ") \
@@ -83,46 +88,9 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
     done
 fi
 
-feat_train_dir=${feats_dir}/${dumpdir}/train; mkdir -p ${feat_train_dir}
-feat_dev_dir=${feats_dir}/${dumpdir}/dev; mkdir -p ${feat_dev_dir}
-feat_test_dir=${feats_dir}/${dumpdir}/test; mkdir -p ${feat_test_dir}
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-    echo "stage 1: Feature Generation"
-    # compute fbank features
-    fbankdir=${feats_dir}/fbank
-    utils/compute_fbank.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --sample_frequency ${sample_frequency} --speed_perturb ${speed_perturb} \
-        ${feats_dir}/data/train ${exp_dir}/exp/make_fbank/train ${fbankdir}/train
-    utils/fix_data_feat.sh ${fbankdir}/train
-    utils/compute_fbank.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --sample_frequency ${sample_frequency} \
-        ${feats_dir}/data/dev ${exp_dir}/exp/make_fbank/dev ${fbankdir}/dev
-    utils/fix_data_feat.sh ${fbankdir}/dev
-    utils/compute_fbank.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --sample_frequency ${sample_frequency} \
-        ${feats_dir}/data/test ${exp_dir}/exp/make_fbank/test ${fbankdir}/test
-    utils/fix_data_feat.sh ${fbankdir}/test
-     
-    # compute global cmvn
-    utils/compute_cmvn.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} \
-        ${fbankdir}/train ${exp_dir}/exp/make_fbank/train
-
-    # apply cmvn 
-    utils/apply_cmvn.sh --cmd "$train_cmd" --nj $nj \
-        ${fbankdir}/train ${fbankdir}/train/cmvn.json ${exp_dir}/exp/make_fbank/train ${feat_train_dir}
-    utils/apply_cmvn.sh --cmd "$train_cmd" --nj $nj \
-        ${fbankdir}/dev ${fbankdir}/train/cmvn.json ${exp_dir}/exp/make_fbank/dev ${feat_dev_dir}
-    utils/apply_cmvn.sh --cmd "$train_cmd" --nj $nj \
-        ${fbankdir}/test ${fbankdir}/train/cmvn.json ${exp_dir}/exp/make_fbank/test ${feat_test_dir}
-    
-    cp ${fbankdir}/train/text ${fbankdir}/train/speech_shape ${fbankdir}/train/text_shape ${feat_train_dir}
-    cp ${fbankdir}/dev/text ${fbankdir}/dev/speech_shape ${fbankdir}/dev/text_shape ${feat_dev_dir}
-    cp ${fbankdir}/test/text ${fbankdir}/test/speech_shape ${fbankdir}/test/text_shape ${feat_test_dir}
-
-    utils/fix_data_feat.sh ${feat_train_dir}
-    utils/fix_data_feat.sh ${feat_dev_dir}
-    utils/fix_data_feat.sh ${feat_test_dir}
-
-    #generate ark list 
-    utils/gen_ark_list.sh --cmd "$train_cmd" --nj $nj ${feat_train_dir} ${fbankdir}/train ${feat_train_dir}
-    utils/gen_ark_list.sh --cmd "$train_cmd" --nj $nj ${feat_dev_dir} ${fbankdir}/dev ${feat_dev_dir}
+    echo "stage 1: Feature and CMVN Generation"
+    utils/compute_cmvn.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} ${feats_dir}/data/${train_set}
 fi
 
 token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt
@@ -135,17 +103,9 @@ if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
     echo "<blank>" > ${token_list}
     echo "<s>" >> ${token_list}
     echo "</s>" >> ${token_list}
-    utils/text2token.py -s 1 -n 1 --space "" ${feats_dir}/data/train/text | cut -f 2- -d" " | tr " " "\n" \
+    utils/text2token.py -s 1 -n 1 --space "" ${feats_dir}/data/$train_set/text | cut -f 2- -d" " | tr " " "\n" \
         | sort | uniq | grep -a -v -e '^\s*$' | awk '{print $0}' >> ${token_list}
-    num_token=$(cat ${token_list} | wc -l)
     echo "<unk>" >> ${token_list}
-    vocab_size=$(cat ${token_list} | wc -l)
-    awk -v v=,${vocab_size} '{print $0v}' ${feat_train_dir}/text_shape > ${feat_train_dir}/text_shape.char
-    awk -v v=,${vocab_size} '{print $0v}' ${feat_dev_dir}/text_shape > ${feat_dev_dir}/text_shape.char
-    mkdir -p ${feats_dir}/asr_stats_fbank_zh_char/train 
-    mkdir -p ${feats_dir}/asr_stats_fbank_zh_char/dev
-    cp ${feat_train_dir}/speech_shape ${feat_train_dir}/text_shape ${feat_train_dir}/text_shape.char ${feats_dir}/asr_stats_fbank_zh_char/train
-    cp ${feat_dev_dir}/speech_shape ${feat_dev_dir}/text_shape ${feat_dev_dir}/text_shape.char ${feats_dir}/asr_stats_fbank_zh_char/dev
 fi
 
 # Training Stage
@@ -172,31 +132,22 @@ if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
             rank=$i
             local_rank=$i
             gpu_id=$(echo $CUDA_VISIBLE_DEVICES | cut -d',' -f$[$i+1])
-            asr_train_paraformer.py \
+            train.py \
+                --task_name asr \
                 --gpu_id $gpu_id \
                 --use_preprocessor true \
                 --token_type char \
                 --token_list $token_list \
-                --train_data_path_and_name_and_type ${feats_dir}/${dumpdir}/${train_set}/${scp},speech,${type} \
-                --train_data_path_and_name_and_type ${feats_dir}/${dumpdir}/${train_set}/text,text,text \
-                --train_data_path_and_name_and_type ${feats_dir}/embeds/${bert_model_name}/${train_set}/embeds.scp,embed,${type} \
-                --train_shape_file ${feats_dir}/asr_stats_fbank_zh_char/${train_set}/speech_shape \
-                --train_shape_file ${feats_dir}/asr_stats_fbank_zh_char/${train_set}/text_shape.char \
-                --train_shape_file ${feats_dir}/embeds/${bert_model_name}/${train_set}/embeds.shape \
-                --valid_data_path_and_name_and_type ${feats_dir}/${dumpdir}/${valid_set}/${scp},speech,${type} \
-                --valid_data_path_and_name_and_type ${feats_dir}/${dumpdir}/${valid_set}/text,text,text \
-                --valid_data_path_and_name_and_type ${feats_dir}/embeds/${bert_model_name}/${valid_set}/embeds.scp,embed,${type} \
-                --valid_shape_file ${feats_dir}/asr_stats_fbank_zh_char/${valid_set}/speech_shape \
-                --valid_shape_file ${feats_dir}/asr_stats_fbank_zh_char/${valid_set}/text_shape.char  \
-                --valid_shape_file ${feats_dir}/embeds/${bert_model_name}/${valid_set}/embeds.shape \
+                --data_dir ${feats_dir}/data \
+                --train_set ${train_set} \
+                --valid_set ${valid_set} \
+                --cmvn_file ${feats_dir}/data/${train_set}/cmvn/cmvn.mvn \
+                --speed_perturb ${speed_perturb} \
                 --resume true \
                 --output_dir ${exp_dir}/exp/${model_dir} \
                 --config $asr_config \
-                --allow_variable_data_keys true \
-                --input_size $feats_dim \
                 --ngpu $gpu_num \
                 --num_worker_count $count \
-                --multiprocessing_distributed true \
                 --dist_init_method $init_method \
                 --dist_world_size $world_size \
                 --dist_rank $rank \

funasr/bin/train.py

@@ -347,6 +347,12 @@ def get_parser():
         default=True,
         help="Apply preprocessing to data or not",
     )
+    parser.add_argument(
+        "--embed_path",
+        type=str,
+        default=None,
+        help="for model which requires embeds",
+    )
 
     # optimization related
     parser.add_argument(

funasr/utils/prepare_data.py

@@ -181,6 +181,11 @@ def prepare_data(args, distributed_option):
             ["{}/{}/wav.scp".format(args.data_dir, args.valid_set), data_names[0], data_types[0]],
             ["{}/{}/text".format(args.data_dir, args.valid_set), data_names[1], data_types[1]]
         ]
+        if args.embed_path is not None:
+            args.train_data_path_and_name_and_type[0].append(
+                "{}/embed/kaldi_ark".format(os.path.join(args.embed_path, args.train_set, "embeds.scp")))
+            args.valid_data_path_and_name_and_type[0].append(
+                "{}/embed/kaldi_ark".format(os.path.join(args.embed_path, args.dev_set, "embeds.scp")))
     else:
         args.train_data_file = os.path.join(args.data_dir, args.train_set, "data.list")
         args.valid_data_file = os.path.join(args.data_dir, args.valid_set, "data.list")