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
atsr
语帆1a6d9d5cc422
diff --git a/.gitignore b/.gitignoreindex d2b4c53b9..1f2a3d1a7 100644--- a/.gitignore+++ b/.gitignore@@ -26,4 +26,5 @@ emotion2vec*GPT-SoVITS*examples/*/*/outputsexamples/*/*/exp+examples/*/*/tmpcmd_readdiff --git a/examples/industrial_data_pretraining/lcbnet/demo.py b/examples/industrial_data_pretraining/lcbnet/demo.pyindex d0870bc8a..602a986d5 100755--- a/examples/industrial_data_pretraining/lcbnet/demo.py+++ b/examples/industrial_data_pretraining/lcbnet/demo.py@@ -10,7 +10,7 @@ model = AutoModel(model="iic/LCB-NET",# example1-res = model.generate(input='["~/.cache/modelscope/hub/iic/LCB-NET/example/asr_example.wav","~/.cache/modelscope/hub/iic/LCB-NET/example/ocr.txt"]',data_type='["sound", "text"]')+res = model.generate(input=("https://www.modelscope.cn/api/v1/models/iic/LCB-NET/repo?Revision=master&FilePath=example/asr_example.wav","https://www.modelscope.cn/api/v1/models/iic/LCB-NET/repo?Revision=master&FilePath=example/ocr.txt"),data_type=("sound", "text"))print(res)diff --git a/examples/industrial_data_pretraining/lcbnet/demo.sh b/examples/industrial_data_pretraining/lcbnet/demo.shindex 825289188..3e04ccd1e 100755--- a/examples/industrial_data_pretraining/lcbnet/demo.sh+++ b/examples/industrial_data_pretraining/lcbnet/demo.sh@@ -1,5 +1,5 @@-file_dir="/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch"-CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"+file_dir="/home/yf352572/.cache/modelscope/hub/iic/LCB-NET/"+CUDA_VISIBLE_DEVICES="0,1"inference_device="cuda"if [ ${inference_device} == "cuda" ]; then@@ -12,7 +12,7 @@ elsedonefi-inference_dir="outputs/slidespeech_dev_beamsearch_new"+inference_dir="outputs/slidespeech_dev"_logdir="${inference_dir}/logdir"echo "inference_dir: ${inference_dir}"@@ -39,11 +39,11 @@ for JOB in $(seq ${nj}); dopython -m funasr.bin.inference \--config-path=${file_dir} \--config-name="config.yaml" \- ++init_param=${file_dir}/model.pb \+ ++init_param=${file_dir}/model.pt \++tokenizer_conf.token_list=${file_dir}/tokens.txt \++input=[${_logdir}/wav.${JOB}.scp,${_logdir}/ocr.${JOB}.txt] \+data_type='["kaldi_ark", "text"]' \- ++tokenizer_conf.bpemodel=${file_dir}/bpe.model \+ ++tokenizer_conf.bpemodel=${file_dir}/bpe.pt \++output_dir="${inference_dir}/${JOB}" \++device="${inference_device}" \++ncpu=1 \diff --git a/examples/industrial_data_pretraining/lcbnet/demo2.sh b/examples/industrial_data_pretraining/lcbnet/demo2.shdeleted file mode 100755index 69df6d16c..000000000--- a/examples/industrial_data_pretraining/lcbnet/demo2.sh+++ /dev/null@@ -1,71 +0,0 @@-file_dir="/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch"-CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"-inference_device="cuda"-test_set="dev_wav"-if [ ${inference_device} == "cuda" ]; then- nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')-else- inference_batch_size=1- CUDA_VISIBLE_DEVICES=""- for JOB in $(seq ${nj}); do- CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"- done-fi--inference_dir="outputs/slidespeech_dev_beamsearch_wav"-_logdir="${inference_dir}/logdir"-echo "inference_dir: ${inference_dir}"--mkdir -p "${_logdir}"-key_file1=${file_dir}/${test_set}/wav.scp-key_file2=${file_dir}/${test_set}/ocr.txt-split_scps1=-split_scps2=-for JOB in $(seq "${nj}"); do- split_scps1+=" ${_logdir}/wav.${JOB}.scp"- split_scps2+=" ${_logdir}/ocr.${JOB}.txt"-done-utils/split_scp.pl "${key_file1}" ${split_scps1}-utils/split_scp.pl "${key_file2}" ${split_scps2}--gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })-for JOB in $(seq ${nj}); do- {- id=$((JOB-1))- gpuid=${gpuid_list_array[$id]}-- export CUDA_VISIBLE_DEVICES=${gpuid}-- python -m funasr.bin.inference \- --config-path=${file_dir} \- --config-name="config.yaml" \- ++init_param=${file_dir}/model.pb \- ++tokenizer_conf.token_list=${file_dir}/tokens.txt \- ++input=[${_logdir}/wav.${JOB}.scp,${_logdir}/ocr.${JOB}.txt] \- +data_type='["sound", "text"]' \- ++tokenizer_conf.bpemodel=${file_dir}/bpe.model \- ++output_dir="${inference_dir}/${JOB}" \- ++device="${inference_device}" \- ++ncpu=1 \- ++disable_log=true &> ${_logdir}/log.${JOB}.txt-- }&-done-wait---mkdir -p ${inference_dir}/1best_recog--for JOB in $(seq "${nj}"); do- cat "${inference_dir}/${JOB}/1best_recog/token" >> "${inference_dir}/1best_recog/token"-done--echo "Computing WER ..."-sed -e 's/ /\t/' -e 's/ //g' -e 's/▁/ /g' -e 's/\t /\t/' ${inference_dir}/1best_recog/token > ${inference_dir}/1best_recog/token.proc-cp ${file_dir}/${test_set}/text ${inference_dir}/1best_recog/token.ref-cp ${file_dir}/${test_set}/ocr.list ${inference_dir}/1best_recog/ocr.list-python utils/compute_wer.py ${inference_dir}/1best_recog/token.ref ${inference_dir}/1best_recog/token.proc ${inference_dir}/1best_recog/token.cer-tail -n 3 ${inference_dir}/1best_recog/token.cer--./run_bwer_recall.sh ${inference_dir}/1best_recog/-tail -n 6 ${inference_dir}/1best_recog/BWER-UWER.results |head -n 5diff --git a/examples/industrial_data_pretraining/lcbnet/demo2_tmp.sh b/examples/industrial_data_pretraining/lcbnet/demo2_tmp.shdeleted file mode 100755index da6ad686d..000000000--- a/examples/industrial_data_pretraining/lcbnet/demo2_tmp.sh+++ /dev/null@@ -1,71 +0,0 @@-file_dir="/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch"-CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"-inference_device="cuda"-test_set="test_wav"-if [ ${inference_device} == "cuda" ]; then- nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')-else- inference_batch_size=1- CUDA_VISIBLE_DEVICES=""- for JOB in $(seq ${nj}); do- CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"- done-fi--inference_dir="outputs/slidespeech_test_beamsearch_wav"-_logdir="${inference_dir}/logdir"-echo "inference_dir: ${inference_dir}"--mkdir -p "${_logdir}"-key_file1=${file_dir}/${test_set}/wav.scp-key_file2=${file_dir}/${test_set}/ocr.txt-split_scps1=-split_scps2=-for JOB in $(seq "${nj}"); do- split_scps1+=" ${_logdir}/wav.${JOB}.scp"- split_scps2+=" ${_logdir}/ocr.${JOB}.txt"-done-utils/split_scp.pl "${key_file1}" ${split_scps1}-utils/split_scp.pl "${key_file2}" ${split_scps2}--gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })-for JOB in $(seq ${nj}); do- {- id=$((JOB-1))- gpuid=${gpuid_list_array[$id]}-- export CUDA_VISIBLE_DEVICES=${gpuid}-- python -m funasr.bin.inference \- --config-path=${file_dir} \- --config-name="config.yaml" \- ++init_param=${file_dir}/model.pb \- ++tokenizer_conf.token_list=${file_dir}/tokens.txt \- ++input=[${_logdir}/wav.${JOB}.scp,${_logdir}/ocr.${JOB}.txt] \- +data_type='["sound", "text"]' \- ++tokenizer_conf.bpemodel=${file_dir}/bpe.model \- ++output_dir="${inference_dir}/${JOB}" \- ++device="${inference_device}" \- ++ncpu=1 \- ++disable_log=true &> ${_logdir}/log.${JOB}.txt-- }&-done-wait---mkdir -p ${inference_dir}/1best_recog--for JOB in $(seq "${nj}"); do- cat "${inference_dir}/${JOB}/1best_recog/token" >> "${inference_dir}/1best_recog/token"-done--echo "Computing WER ..."-sed -e 's/ /\t/' -e 's/ //g' -e 's/▁/ /g' -e 's/\t /\t/' ${inference_dir}/1best_recog/token > ${inference_dir}/1best_recog/token.proc-cp ${file_dir}/${test_set}/text ${inference_dir}/1best_recog/token.ref-cp ${file_dir}/${test_set}/ocr.list ${inference_dir}/1best_recog/ocr.list-python utils/compute_wer.py ${inference_dir}/1best_recog/token.ref ${inference_dir}/1best_recog/token.proc ${inference_dir}/1best_recog/token.cer-tail -n 3 ${inference_dir}/1best_recog/token.cer--./run_bwer_recall.sh ${inference_dir}/1best_recog/-tail -n 6 ${inference_dir}/1best_recog/BWER-UWER.results |head -n 5diff --git a/examples/industrial_data_pretraining/lcbnet/demo_pdb.sh b/examples/industrial_data_pretraining/lcbnet/demo_pdb.shdeleted file mode 100755index 0747a8d7b..000000000--- a/examples/industrial_data_pretraining/lcbnet/demo_pdb.sh+++ /dev/null@@ -1,18 +0,0 @@-file_dir="/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch"--#CUDA_VISIBLE_DEVICES="" \-python -m funasr.bin.inference \---config-path=${file_dir} \---config-name="config.yaml" \-++init_param=${file_dir}/model.pb \-++tokenizer_conf.token_list=${file_dir}/tokens.txt \-+input=["${file_dir}/example/asr_example.wav","${file_dir}/example/ocr.txt"] \-+data_type='["sound","text"]' \-++tokenizer_conf.bpemodel=${file_dir}/bpe.model \-++output_dir="./outputs/debug" \-++device="cpu" \--#++input=["/nfs/yufan.yf/workspace/espnet/egs2/youtube_ppt/asr/dump/raw/dev_oracle_v1_new/data/format.1/YTB+--tMoLpQI-w+00322.wav"] \-#+data_type='["sound"]' \-#++input=["/nfs/yufan.yf/workspace/espnet/egs2/youtube_ppt/asr/dump/raw/dev_oracle_v1_new/data/format.1/YTB+--tMoLpQI-w+00322.wav","/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch/example/ocr2.txt"] \-#+data_type='["sound","text"]' \diff --git a/examples/industrial_data_pretraining/lcbnet/demo_pdb2.sh b/examples/industrial_data_pretraining/lcbnet/demo_pdb2.shdeleted file mode 100755index 557e9b2d8..000000000--- a/examples/industrial_data_pretraining/lcbnet/demo_pdb2.sh+++ /dev/null@@ -1,15 +0,0 @@-file_dir="/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch"--#CUDA_VISIBLE_DEVICES="" \-python -m funasr.bin.inference \---config-path=${file_dir} \---config-name="config.yaml" \-++init_param=${file_dir}/model.pb \-++tokenizer_conf.token_list=${file_dir}/tokens.txt \-++input=[${file_dir}/dev_wav/wav.scp,${file_dir}/dev_wav/ocr.txt] \-+data_type='["sound", "text"]' \-++tokenizer_conf.bpemodel=${file_dir}/bpe.model \-++output_dir="./outputs/debug" \-++device="cpu" \--#++input=[${file_dir}/dev_wav/wav.scp,${file_dir}/dev_wav/ocr.txt] \diff --git a/examples/industrial_data_pretraining/lcbnet/demo_tmp1.sh b/examples/industrial_data_pretraining/lcbnet/demo_tmp1.shdeleted file mode 100755index 488f7d2a1..000000000--- a/examples/industrial_data_pretraining/lcbnet/demo_tmp1.sh+++ /dev/null@@ -1,71 +0,0 @@-file_dir="/nfs/yufan.yf/workspace/github/FunASR/examples/industrial_data_pretraining/lcbnet/exp/speech_lcbnet_contextual_asr-en-16k-bpe-vocab5002-pytorch"-CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"-inference_device="cuda"--if [ ${inference_device} == "cuda" ]; then- nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')-else- inference_batch_size=1- CUDA_VISIBLE_DEVICES=""- for JOB in $(seq ${nj}); do- CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"- done-fi--inference_dir="outputs/slidespeech_test_beamsearch_new"-_logdir="${inference_dir}/logdir"-echo "inference_dir: ${inference_dir}"--mkdir -p "${_logdir}"-key_file1=${file_dir}/test/wav.scp-key_file2=${file_dir}/test/ocr.txt-split_scps1=-split_scps2=-for JOB in $(seq "${nj}"); do- split_scps1+=" ${_logdir}/wav.${JOB}.scp"- split_scps2+=" ${_logdir}/ocr.${JOB}.txt"-done-utils/split_scp.pl "${key_file1}" ${split_scps1}-utils/split_scp.pl "${key_file2}" ${split_scps2}--gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })-for JOB in $(seq ${nj}); do- {- id=$((JOB-1))- gpuid=${gpuid_list_array[$id]}-- export CUDA_VISIBLE_DEVICES=${gpuid}-- python -m funasr.bin.inference \- --config-path=${file_dir} \- --config-name="config.yaml" \- ++init_param=${file_dir}/model.pb \- ++tokenizer_conf.token_list=${file_dir}/tokens.txt \- ++input=[${_logdir}/wav.${JOB}.scp,${_logdir}/ocr.${JOB}.txt] \- +data_type='["kaldi_ark", "text"]' \- ++tokenizer_conf.bpemodel=${file_dir}/bpe.model \- ++output_dir="${inference_dir}/${JOB}" \- ++device="${inference_device}" \- ++ncpu=1 \- ++disable_log=true &> ${_logdir}/log.${JOB}.txt-- }&-done-wait---mkdir -p ${inference_dir}/1best_recog--for JOB in $(seq "${nj}"); do- cat "${inference_dir}/${JOB}/1best_recog/token" >> "${inference_dir}/1best_recog/token"-done--echo "Computing WER ..."-sed -e 's/ /\t/' -e 's/ //g' -e 's/▁/ /g' -e 's/\t /\t/' ${inference_dir}/1best_recog/token > ${inference_dir}/1best_recog/token.proc-cp ${file_dir}/test/text ${inference_dir}/1best_recog/token.ref-cp ${file_dir}/test/ocr.list ${inference_dir}/1best_recog/ocr.list-python utils/compute_wer.py ${inference_dir}/1best_recog/token.ref ${inference_dir}/1best_recog/token.proc ${inference_dir}/1best_recog/token.cer-tail -n 3 ${inference_dir}/1best_recog/token.cer--./run_bwer_recall.sh ${inference_dir}/1best_recog/-tail -n 6 ${inference_dir}/1best_recog/BWER-UWER.results |head -n 5diff --git a/funasr/utils/load_utils.py b/funasr/utils/load_utils.pyindex 644af2324..84c38f9b9 100644--- a/funasr/utils/load_utils.py+++ b/funasr/utils/load_utils.py@@ -89,8 +89,6 @@ def load_bytes(input):return arraydef extract_fbank(data, data_len = None, data_type: str="sound", frontend=None, **kwargs):- # import pdb;- # pdb.set_trace()if isinstance(data, np.ndarray):data = torch.from_numpy(data)if len(data.shape) < 2: