Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
bugfix
游雁cab0af88d8d6
diff --git a/examples/aishell/branchformer/infer.sh b/examples/aishell/branchformer/infer.shnew file mode 100644index 000000000..bcd4305f6--- /dev/null+++ b/examples/aishell/branchformer/infer.sh@@ -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" \+diff --git a/examples/aishell/branchformer/run.sh b/examples/aishell/branchformer/run.shindex dba8b96e2..a8c655ebb 100755--- a/examples/aishell/branchformer/run.sh+++ b/examples/aishell/branchformer/run.sh@@ -43,6 +43,7 @@ config=branchformer_12e_6d_2048_256.yamlmodel_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; thenecho "stage -1: Data Download"mkdir -p ${raw_data}@@ -85,7 +86,7 @@ echo "dictionary: ${token_list}"if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; thenecho "stage 2: Dictionary Preparation"mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/-+echo "make a dictionary"echo "<blank>" > ${token_list}echo "<s>" >> ${token_list}@@ -130,7 +131,7 @@ fiif [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenecho "stage 5: Inference"- if ${inference_device} == "cuda"; then+ if [ ${inference_device} == "cuda" ]; thennj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')elseinference_batch_size=1@@ -142,7 +143,7 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenfor 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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneutils/split_scp.pl "${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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then++input="${_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@@ -187,10 +188,10 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneecho "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.refpython utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.certail -n 3 ${inference_dir}/1best_recog/text.cerdone-fi+fi\ No newline at end of filediff --git a/examples/aishell/conformer/infer.sh b/examples/aishell/conformer/infer.shindex a64df547e..bcd4305f6 100644--- a/examples/aishell/conformer/infer.sh+++ b/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 @@ python funasr/bin/inference.py \++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" \diff --git a/examples/aishell/conformer/run.sh b/examples/aishell/conformer/run.shindex 3029cbd07..7d6a4f76b 100755--- a/examples/aishell/conformer/run.sh+++ b/examples/aishell/conformer/run.sh@@ -14,10 +14,10 @@ stop_stage=5# feature configurationnj=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# dataraw_data=../raw_data@@ -142,7 +142,7 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenfor 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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneutils/split_scp.pl "${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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneecho "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.refpython utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.certail -n 3 ${inference_dir}/1best_recog/text.cerdonediff --git a/examples/aishell/e_branchformer/infer.sh b/examples/aishell/e_branchformer/infer.shnew file mode 100644index 000000000..bcd4305f6--- /dev/null+++ b/examples/aishell/e_branchformer/infer.sh@@ -0,0 +1,12 @@+++python -m funasr.bin.inference \+--config-path="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3" \+--config-name="config.yaml" \+++init_param="/mnt/workspace/FunASR/examples/aishell/paraformer/exp/baseline_paraformer_conformer_12e_6d_2048_256_zh_char_exp3/model.pt.ep38" \+++tokenizer_conf.token_list="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/zh_token_list/char/tokens.txt" \+++frontend_conf.cmvn_file="/mnt/nfs/zhifu.gzf/data/AISHELL-1-feats/DATA/data/train/am.mvn" \+++input="/mnt/nfs/zhifu.gzf/data/AISHELL-1/data_aishell/wav/train/S0002/BAC009S0002W0122.wav" \+++output_dir="./outputs/debug" \+++device="cuda:0" \+diff --git a/examples/aishell/e_branchformer/run.sh b/examples/aishell/e_branchformer/run.shindex ed765b872..c7cae8757 100755--- a/examples/aishell/e_branchformer/run.sh+++ b/examples/aishell/e_branchformer/run.sh@@ -43,6 +43,7 @@ config=e_branchformer_12e_6d_2048_256.yamlmodel_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; thenecho "stage -1: Data Download"mkdir -p ${raw_data}@@ -85,7 +86,7 @@ echo "dictionary: ${token_list}"if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; thenecho "stage 2: Dictionary Preparation"mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/-+echo "make a dictionary"echo "<blank>" > ${token_list}echo "<s>" >> ${token_list}@@ -130,7 +131,7 @@ fiif [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenecho "stage 5: Inference"- if ${inference_device} == "cuda"; then+ if [ ${inference_device} == "cuda" ]; thennj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')elseinference_batch_size=1@@ -142,7 +143,7 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenfor 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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneutils/split_scp.pl "${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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then++input="${_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@@ -187,10 +188,10 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneecho "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.refpython utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.certail -n 3 ${inference_dir}/1best_recog/text.cerdone-fi+fi\ No newline at end of filediff --git a/examples/aishell/paraformer/infer.sh b/examples/aishell/paraformer/infer.shnew file mode 100644index 000000000..bcd4305f6--- /dev/null+++ b/examples/aishell/paraformer/infer.sh@@ -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" \+diff --git a/examples/aishell/paraformer/run.sh b/examples/aishell/paraformer/run.shindex d6f1a2cbc..0f41efe7f 100755--- a/examples/aishell/paraformer/run.sh+++ b/examples/aishell/paraformer/run.sh@@ -130,7 +130,7 @@ fiif [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenecho "stage 5: Inference"- if ${inference_device} == "cuda"; then+ if [ ${inference_device} == "cuda" ]; thennj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')elseinference_batch_size=1@@ -142,7 +142,7 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenfor 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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneutils/split_scp.pl "${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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then++input="${_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@@ -187,10 +187,10 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneecho "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.refpython utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.certail -n 3 ${inference_dir}/1best_recog/text.cerdone-fi+fi\ No newline at end of filediff --git a/examples/aishell/paraformer/utils/postprocess_text_zh.py b/examples/aishell/paraformer/utils/postprocess_text_zh.pynew file mode 100755index 000000000..9e517a4e1--- /dev/null+++ b/examples/aishell/paraformer/utils/postprocess_text_zh.py@@ -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)diff --git a/examples/aishell/transformer/infer.sh b/examples/aishell/transformer/infer.shnew file mode 100644index 000000000..bcd4305f6--- /dev/null+++ b/examples/aishell/transformer/infer.sh@@ -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" \+diff --git a/examples/aishell/transformer/run.sh b/examples/aishell/transformer/run.shindex 3a0216168..c0ff4338f 100755--- a/examples/aishell/transformer/run.sh+++ b/examples/aishell/transformer/run.sh@@ -43,6 +43,7 @@ config=transformer_12e_6d_2048_256.yamlmodel_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; thenecho "stage -1: Data Download"mkdir -p ${raw_data}@@ -85,7 +86,7 @@ echo "dictionary: ${token_list}"if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; thenecho "stage 2: Dictionary Preparation"mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/-+echo "make a dictionary"echo "<blank>" > ${token_list}echo "<s>" >> ${token_list}@@ -142,7 +143,7 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thenfor 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 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then++input="${_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@@ -187,10 +188,10 @@ if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; thendoneecho "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.refpython utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.certail -n 3 ${inference_dir}/1best_recog/text.cerdone-fi+fi\ No newline at end of filediff --git a/funasr/train_utils/trainer.py b/funasr/train_utils/trainer.pyindex 4b85a66a6..61b9004f4 100644--- a/funasr/train_utils/trainer.py+++ b/funasr/train_utils/trainer.py@@ -109,12 +109,8 @@ class Trainer: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):"""