bugfix

游雁

2024-02-21 cab0af88d8d671d7a358dbdd1c9645cc4cbafd1d

bugfix

 examples/aishell/branchformer/infer.sh

New file
@@ -0,0 +1,12 @@


python -m funasr.bin.inference \
--config-path="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3" \
--config-name="config.yaml" \
++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \
++tokenizer_conf.token_list="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/zh_token_list/char/tokens.txt" \
++frontend_conf.cmvn_file="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/train/am.mvn" \
++input="/mnt/nfs/zhifu.gzf/data/AISHELL-1/data_aishell/wav/train/S0002/BAC009S0002W0122.wav" \
++output_dir="./outputs/debug" \
++device="cuda:0" \


 examples/aishell/branchformer/run.sh

@@ -43,6 +43,7 @@
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"



if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
    echo "stage -1: Data Download"
    mkdir -p ${raw_data}
@@ -130,7 +131,7 @@
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
  echo "stage 5: Inference"

  if ${inference_device} == "cuda"; then
  if [ ${inference_device} == "cuda" ]; then
      nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  else
      inference_batch_size=1
@@ -142,7 +143,7 @@

  for dset in ${test_sets}; do

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

    mkdir -p "${_logdir}"
@@ -155,7 +156,7 @@
    done
    utils/split_scp.pl "${key_file}" ${split_scps}

    gpuid_list_array=(${gpuid_list//,/ })
    gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
    for JOB in $(seq ${nj}); do
        {
          id=$((JOB-1))
@@ -171,7 +172,7 @@
          ++input="${_logdir}/keys.${JOB}.scp" \
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++batch_size="${inference_batch_size}"
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&

    done
@@ -187,8 +188,8 @@
    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

 examples/aishell/conformer/infer.sh

@@ -1,6 +1,6 @@


python funasr/bin/inference.py \
python -m funasr.bin.inference \
--config-path="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3" \
--config-name="config.yaml" \
++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \
@@ -8,5 +8,5 @@
++frontend_conf.cmvn_file="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/train/am.mvn" \
++input="/mnt/nfs/zhifu.gzf/data/AISHELL-1/data_aishell/wav/train/S0002/BAC009S0002W0122.wav" \
++output_dir="./outputs/debug" \
++device="cpu" \
++device="cuda:0" \


 examples/aishell/conformer/run.sh

@@ -14,10 +14,10 @@
# feature configuration
nj=32

inference_device="cuda" #"cpu"
inference_device="cuda" #"cpu", "cuda:0", "cuda:1"
inference_checkpoint="model.pt"
inference_scp="wav.scp"
inference_batch_size=32
inference_batch_size=1

# data
raw_data=../raw_data
@@ -142,7 +142,7 @@

  for dset in ${test_sets}; do

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

    mkdir -p "${_logdir}"
@@ -155,7 +155,7 @@
    done
    utils/split_scp.pl "${key_file}" ${split_scps}

    gpuid_list_array=(${gpuid_list//,/ })
    gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
    for JOB in $(seq ${nj}); do
        {
          id=$((JOB-1))
@@ -187,8 +187,8 @@
    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

 examples/aishell/e_branchformer/infer.sh

New file
@@ -0,0 +1,12 @@


python -m funasr.bin.inference \
--config-path="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3" \
--config-name="config.yaml" \
++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \
++tokenizer_conf.token_list="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/zh_token_list/char/tokens.txt" \
++frontend_conf.cmvn_file="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/train/am.mvn" \
++input="/mnt/nfs/zhifu.gzf/data/AISHELL-1/data_aishell/wav/train/S0002/BAC009S0002W0122.wav" \
++output_dir="./outputs/debug" \
++device="cuda:0" \


 examples/aishell/e_branchformer/run.sh

@@ -43,6 +43,7 @@
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"



if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
    echo "stage -1: Data Download"
    mkdir -p ${raw_data}
@@ -130,7 +131,7 @@
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
  echo "stage 5: Inference"

  if ${inference_device} == "cuda"; then
  if [ ${inference_device} == "cuda" ]; then
      nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  else
      inference_batch_size=1
@@ -142,7 +143,7 @@

  for dset in ${test_sets}; do

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

    mkdir -p "${_logdir}"
@@ -155,7 +156,7 @@
    done
    utils/split_scp.pl "${key_file}" ${split_scps}

    gpuid_list_array=(${gpuid_list//,/ })
    gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
    for JOB in $(seq ${nj}); do
        {
          id=$((JOB-1))
@@ -171,7 +172,7 @@
          ++input="${_logdir}/keys.${JOB}.scp" \
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++batch_size="${inference_batch_size}"
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&

    done
@@ -187,8 +188,8 @@
    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

 examples/aishell/paraformer/infer.sh

New file
@@ -0,0 +1,12 @@


python -m funasr.bin.inference \
--config-path="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3" \
--config-name="config.yaml" \
++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \
++tokenizer_conf.token_list="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/zh_token_list/char/tokens.txt" \
++frontend_conf.cmvn_file="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/train/am.mvn" \
++input="/mnt/nfs/zhifu.gzf/data/AISHELL-1/data_aishell/wav/train/S0002/BAC009S0002W0122.wav" \
++output_dir="./outputs/debug" \
++device="cuda:0" \


 examples/aishell/paraformer/run.sh

@@ -130,7 +130,7 @@
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
  echo "stage 5: Inference"

  if ${inference_device} == "cuda"; then
  if [ ${inference_device} == "cuda" ]; then
      nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
  else
      inference_batch_size=1
@@ -142,7 +142,7 @@

  for dset in ${test_sets}; do

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

    mkdir -p "${_logdir}"
@@ -155,7 +155,7 @@
    done
    utils/split_scp.pl "${key_file}" ${split_scps}

    gpuid_list_array=(${gpuid_list//,/ })
    gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
    for JOB in $(seq ${nj}); do
        {
          id=$((JOB-1))
@@ -171,7 +171,7 @@
          ++input="${_logdir}/keys.${JOB}.scp" \
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++batch_size="${inference_batch_size}"
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&

    done
@@ -187,8 +187,8 @@
    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

 examples/aishell/paraformer/utils/postprocess_text_zh.py

New file
@@ -0,0 +1,31 @@

import sys
import re

in_f = sys.argv[1]
out_f = sys.argv[2]


with open(in_f, "r", encoding="utf-8") as f:
  lines = f.readlines()

with open(out_f, "w", encoding="utf-8") as f:
  for line in lines:
    outs = line.strip().split(" ", 1)
    if len(outs) == 2:
      idx, text = outs
      text = re.sub("</s>", "", text)
      text = re.sub("<s>", "", text)
      text = re.sub("@@", "", text)
      text = re.sub("@", "", text)
      text = re.sub("<unk>", "", text)
      text = re.sub(" ", "", text)
      text = text.lower()
    else:
      idx = outs[0]
      text = " "

    text = [x for x in text]
    text = " ".join(text)
    out = "{} {}\n".format(idx, text)
    f.write(out)

 examples/aishell/transformer/infer.sh

New file
@@ -0,0 +1,12 @@


python -m funasr.bin.inference \
--config-path="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3" \
--config-name="config.yaml" \
++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \
++tokenizer_conf.token_list="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/zh_token_list/char/tokens.txt" \
++frontend_conf.cmvn_file="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/train/am.mvn" \
++input="/mnt/nfs/zhifu.gzf/data/AISHELL-1/data_aishell/wav/train/S0002/BAC009S0002W0122.wav" \
++output_dir="./outputs/debug" \
++device="cuda:0" \


 examples/aishell/transformer/run.sh

@@ -43,6 +43,7 @@
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"



if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
    echo "stage -1: Data Download"
    mkdir -p ${raw_data}
@@ -142,7 +143,7 @@

  for dset in ${test_sets}; do

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

    mkdir -p "${_logdir}"
@@ -171,7 +172,7 @@
          ++input="${_logdir}/keys.${JOB}.scp" \
          ++output_dir="${inference_dir}/${JOB}" \
          ++device="${inference_device}" \
          ++batch_size="${inference_batch_size}"
          ++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
        }&

    done
@@ -187,8 +188,8 @@
    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

 funasr/train_utils/trainer.py

@@ -109,12 +109,8 @@
        
        print(f'Checkpoint saved to {filename}')
        latest = Path(os.path.join(self.output_dir, f'model.pt'))
        try:
            latest.unlink()
        except:
            pass
        torch.save(state, latest)

        latest.symlink_to(filename)
    
    def _resume_checkpoint(self, resume_path):
        """