update repo

嘉渊

2023-05-11 4e37a5fda20f0878b593b8ba2b9ea46db63743b5

 egs/librispeech_100h/conformer/run.sh

@@ -3,8 +3,8 @@
. ./path.sh || exit 1;

# machines configuration
CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
gpu_num=8
CUDA_VISIBLE_DEVICES="0,1"
gpu_num=2
count=1
gpu_inference=true  # Whether to perform gpu decoding, set false for cpu decoding
# for gpu decoding, inference_nj=ngpu*njob; for cpu decoding, inference_nj=njob
@@ -16,30 +16,26 @@
feats_dir="../DATA" #feature output dictionary
exp_dir="."
lang=en
dumpdir=dump/fbank
feats_type=fbank
token_type=bpe
dataset_type=large
scp=feats.scp
type=kaldi_ark
stage=3
stop_stage=4
type=sound
scp=wav.scp
stage=1
stop_stage=1

# feature configuration
feats_dim=80
sample_frequency=16000
nj=100
speed_perturb="0.9,1.0,1.1"
nj=64

# data
data_librispeech=
raw_data=
data_url=www.openslr.org/resources/12

# bpe model
nbpe=5000
bpemode=unigram

# exp tag
tag=""
tag="exp1"

. utils/parse_options.sh || exit 1;

@@ -49,13 +45,12 @@
set -u
set -o pipefail

train_set=train_960
train_set=train_clean_100
valid_set=dev
test_sets="test_clean test_other dev_clean dev_other"

asr_config=conf/train_asr_conformer.yaml
#asr_config=conf/train_asr_conformer_uttnorm.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.yaml
#inference_config=conf/decode_asr_transformer_beam60_ctc0.3.yaml
@@ -73,69 +68,25 @@
    _ngpu=0
fi

if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
    echo "stage 0: Data preparation"
    # Data preparation
    for x in train-clean-100 train-clean-360 train-other-500 dev-clean dev-other test-clean test-other; do
        local/data_prep_librispeech.sh ${data_librispeech}/LibriSpeech/${x} ${feats_dir}/data/${x//-/_}

if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
    echo "stage -1: Data Download"
    for part in dev-clean test-clean dev-other test-other train-clean-100; do
        local/download_and_untar.sh ${raw_data} ${data_url} ${part}
    done
fi

feat_train_dir=${feats_dir}/${dumpdir}/$train_set; mkdir -p ${feat_train_dir}
feat_dev_clean_dir=${feats_dir}/${dumpdir}/dev_clean; mkdir -p ${feat_dev_clean_dir}
feat_dev_other_dir=${feats_dir}/${dumpdir}/dev_other; mkdir -p ${feat_dev_other_dir}
feat_test_clean_dir=${feats_dir}/${dumpdir}/test_clean; mkdir -p ${feat_test_clean_dir}
feat_test_other_dir=${feats_dir}/${dumpdir}/test_other; mkdir -p ${feat_test_other_dir}
feat_dev_dir=${feats_dir}/${dumpdir}/$valid_set; mkdir -p ${feat_dev_dir}
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
    echo "stage 0: Data preparation"
    # Data preparation
    for x in dev-clean dev-other test-clean test-other train-clean-100; do
        local/data_prep.sh ${raw_data}/LibriSpeech/${x} ${feats_dir}/data/${x//-/_}
    done
fi

if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
    echo "stage 1: Feature Generation"
    # compute fbank features
    fbankdir=${feats_dir}/fbank
    for x in dev_clean dev_other test_clean test_other; do
        utils/compute_fbank.sh --cmd "$train_cmd" --nj 1 --max_lengths 3000 --feats_dim ${feats_dim} --sample_frequency ${sample_frequency} \
            ${feats_dir}/data/${x} ${exp_dir}/exp/make_fbank/${x} ${fbankdir}/${x}
        utils/fix_data_feat.sh ${fbankdir}/${x}
    done

    mkdir ${feats_dir}/data/$train_set
    train_sets="train_clean_100 train_clean_360 train_other_500"
    for file in wav.scp text; do
        ( for f in $train_sets; do cat $feats_dir/data/$f/$file; done ) | sort -k1 > $feats_dir/data/$train_set/$file || exit 1;
    done
    utils/compute_fbank.sh --cmd "$train_cmd" --nj $nj --max_lengths 3000 --feats_dim ${feats_dim} --sample_frequency ${sample_frequency} --speed_perturb ${speed_perturb} \
    ${feats_dir}/data/$train_set ${exp_dir}/exp/make_fbank/$train_set ${fbankdir}/$train_set
    utils/fix_data_feat.sh ${fbankdir}/$train_set

    # compute global cmvn
    utils/compute_cmvn.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} \
        ${fbankdir}/$train_set ${exp_dir}/exp/make_fbank/$train_set

    # apply cmvn
    utils/apply_cmvn.sh --cmd "$train_cmd" --nj $nj \
        ${fbankdir}/$train_set ${fbankdir}/$train_set/cmvn.json ${exp_dir}/exp/make_fbank/$train_set ${feat_train_dir}
    utils/apply_cmvn.sh --cmd "$train_cmd" --nj 1 \
        ${fbankdir}/dev_clean ${fbankdir}/$train_set/cmvn.json ${exp_dir}/exp/make_fbank/dev_clean ${feat_dev_clean_dir}
    utils/apply_cmvn.sh --cmd "$train_cmd" --nj 1\
        ${fbankdir}/dev_other ${fbankdir}/$train_set/cmvn.json ${exp_dir}/exp/make_fbank/dev_other ${feat_dev_other_dir}
    utils/apply_cmvn.sh --cmd "$train_cmd" --nj 1 \
        ${fbankdir}/test_clean ${fbankdir}/$train_set/cmvn.json ${exp_dir}/exp/make_fbank/test_clean ${feat_test_clean_dir}
    utils/apply_cmvn.sh --cmd "$train_cmd" --nj 1 \
        ${fbankdir}/test_other ${fbankdir}/$train_set/cmvn.json ${exp_dir}/exp/make_fbank/test_other ${feat_test_other_dir}

    cp ${fbankdir}/$train_set/text ${fbankdir}/$train_set/speech_shape ${fbankdir}/$train_set/text_shape ${feat_train_dir}
    cp ${fbankdir}/dev_clean/text ${fbankdir}/dev_clean/speech_shape ${fbankdir}/dev_clean/text_shape ${feat_dev_clean_dir}
    cp ${fbankdir}/dev_other/text ${fbankdir}/dev_other/speech_shape ${fbankdir}/dev_other/text_shape ${feat_dev_other_dir}
    cp ${fbankdir}/test_clean/text ${fbankdir}/test_clean/speech_shape ${fbankdir}/test_clean/text_shape ${feat_test_clean_dir}
    cp ${fbankdir}/test_other/text ${fbankdir}/test_other/speech_shape ${fbankdir}/test_other/text_shape ${feat_test_other_dir}

    dev_sets="dev_clean dev_other"
    for file in feats.scp text speech_shape text_shape; do
        ( for f in $dev_sets; do cat $feats_dir/${dumpdir}/$f/$file; done ) | sort -k1 > $feat_dev_dir/$file || exit 1;
    done

    #generate ark list
    utils/gen_ark_list.sh --cmd "$train_cmd" --nj $nj ${feat_train_dir} ${fbankdir}/${train_set} ${feat_train_dir}
    utils/gen_ark_list.sh --cmd "$train_cmd" --nj $nj ${feat_dev_dir} ${fbankdir}/${valid_set} ${feat_dev_dir}
    echo "stage 1: Feature and CMVN Generation"
    utils/compute_cmvn.sh --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} ${feats_dir}/data/${train_set}
fi

dict=${feats_dir}/data/lang_char/${train_set}_${bpemode}${nbpe}_units.txt
@@ -152,15 +103,6 @@
    spm_train --input=${feats_dir}/data/lang_char/input.txt --vocab_size=${nbpe} --model_type=${bpemode} --model_prefix=${bpemodel} --input_sentence_size=100000000
    spm_encode --model=${bpemodel}.model --output_format=piece < ${feats_dir}/data/lang_char/input.txt | tr ' ' '\n' | sort | uniq | awk '{print $0}' >> ${dict}
    echo "<unk>" >> ${dict}
    wc -l ${dict}

    vocab_size=$(cat ${dict} | wc -l)
    awk -v v=,${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_set
    mkdir -p ${feats_dir}/asr_stats_fbank_zh_char/$valid_set
    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_set
    cp ${feat_dev_dir}/speech_shape ${feat_dev_dir}/text_shape ${feat_dev_dir}/text_shape.char ${feats_dir}/asr_stats_fbank_zh_char/$valid_set
fi