test

语帆

2024-02-28 eb92e79fb94e7b3df8f27c8ce3e607a70dff2a2e

 examples/aishell/paraformer/run.sh

@@ -1,12 +1,11 @@
#!/usr/bin/env bash



CUDA_VISIBLE_DEVICES="0,1"

# general configuration
feats_dir="../DATA" #feature output dictionary
exp_dir="."
exp_dir=`pwd`
lang=zh
token_type=char
stage=0
@@ -18,6 +17,7 @@
inference_device="cuda" #"cpu"
inference_checkpoint="model.pt"
inference_scp="wav.scp"
inference_batch_size=32

# data
raw_data=../raw_data
@@ -39,7 +39,7 @@
valid_set=dev
test_sets="dev test"

config=train_asr_paraformer_conformer_12e_6d_2048_256.yaml
config=paraformer_conformer_12e_6d_2048_256.yaml
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"


@@ -105,10 +105,12 @@
  echo "stage 4: ASR Training"

  mkdir -p ${exp_dir}/exp/${model_dir}
  log_file="${exp_dir}/exp/${model_dir}/train.log.txt"
  current_time=$(date "+%Y-%m-%d_%H-%M")
  log_file="${exp_dir}/exp/${model_dir}/train.log.txt.${current_time}"
  echo "log_file: ${log_file}"

  gpu_num=$(echo CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  export CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES
  gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  torchrun \
  --nnodes 1 \
  --nproc_per_node ${gpu_num} \
@@ -128,21 +130,21 @@
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
  echo "stage 5: Inference"

  if ${inference_device} == "cuda"; then
      nj=$(echo CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  if [ ${inference_device} == "cuda" ]; then
      nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  else
      nj=$njob
      batch_size=1
      gpuid_list=""
      inference_batch_size=1
      CUDA_VISIBLE_DEVICES=""
      for JOB in $(seq ${nj}); do
          gpuid_list=CUDA_VISIBLE_DEVICES"-1,"
          CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"
      done
  fi

  for dset in ${test_sets}; do

    inference_dir="${asr_exp}/${inference_checkpoint}/${dset}"
    inference_dir="${exp_dir}/exp/${model_dir}/inference-${inference_checkpoint}/${dset}"
    _logdir="${inference_dir}/logdir"
    echo "inference_dir: ${inference_dir}"

    mkdir -p "${_logdir}"
    data_dir="${feats_dir}/data/${dset}"
@@ -154,8 +156,13 @@
    done
    utils/split_scp.pl "${key_file}" ${split_scps}

    gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
    for JOB in $(seq ${nj}); do
        {
          id=$((JOB-1))
          gpuid=${gpuid_list_array[$id]}

          export CUDA_VISIBLE_DEVICES=${gpuid}
          python ../../../funasr/bin/inference.py \
          --config-path="${exp_dir}/exp/${model_dir}" \
          --config-name="config.yaml" \
@@ -164,7 +171,10 @@
          ++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
          ++input="${_logdir}/keys.${JOB}.scp" \
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}"
          ++device="${inference_device}" \
          ++ncpu=1 \
          ++disable_log=true \
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&

    done
@@ -180,10 +190,10 @@
    done

    echo "Computing WER ..."
    cp ${inference_dir}/1best_recog/text ${inference_dir}/1best_recog/text.proc
    cp ${data_dir}/text ${inference_dir}/1best_recog/text.ref
    python utils/postprocess_text_zh.py ${inference_dir}/1best_recog/text ${inference_dir}/1best_recog/text.proc
    python utils/postprocess_text_zh.py  ${data_dir}/text ${inference_dir}/1best_recog/text.ref
    python utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.cer
    tail -n 3 ${inference_dir}/1best_recog/text.cer
  done

fi
fi