update repo

嘉渊

2023-05-15 688fb902dd625981060b00788ed70c4c155d2b50

update repo

 egs/aishell/paraformerbert/local/extract_embeds.sh

@@ -5,8 +5,6 @@

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 @@

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 @@
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 @@

# 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 @@
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 @@
    _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 @@
    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 @@
    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 @@
            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 @@
        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 @@
            ["{}/{}/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")