aishell example

examples/aishell/paraformer/run.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 
-. ./path.sh || exit 1;
+workspace=`pwd`
 
 # machines configuration
 CUDA_VISIBLE_DEVICES="0,1"
@@ -39,7 +39,7 @@ train_set=train
 valid_set=dev
 test_sets="dev test"
 
-asr_config=conf/train_asr_paraformer_conformer_12e_6d_2048_256.yaml
+asr_config=train_asr_paraformer_conformer_12e_6d_2048_256.yaml
 model_dir="baseline_$(basename "${asr_config}" .yaml)_${lang}_${token_type}_${tag}"
 
 #inference_config=conf/decode_asr_transformer_noctc_1best.yaml
@@ -74,19 +74,21 @@ if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
         utils/text2token.py -n 1 -s 1 ${feats_dir}/data/${x}/text > ${feats_dir}/data/${x}/text.org
         mv ${feats_dir}/data/${x}/text.org ${feats_dir}/data/${x}/text
 
-        python funasr/datasets/audio_datasets/scp2jsonl.py \
-        ++scp_file_list='["${feats_dir}/data/${x}/wav.scp", "${feats_dir}/data/${x}/text"]' \
+        # convert wav.scp text to jsonl
+        scp_file_list_arg="++scp_file_list='[\"${feats_dir}/data/${x}/wav.scp\",\"${feats_dir}/data/${x}/text\"]'"
+        python ../../../funasr/datasets/audio_datasets/scp2jsonl.py \
         ++data_type_list='["source", "target"]' \
-        ++jsonl_file_out=${feats_dir}/data/${x}/audio_datasets.jsonl
+        ++jsonl_file_out=${feats_dir}/data/${x}/audio_datasets.jsonl \
+        ${scp_file_list_arg}
     done
 fi
 
 if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
     echo "stage 1: Feature and CMVN Generation"
 #    utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0
-    python funasr/bin/compute_audio_cmvn.py \
-    --config-path "/Users/zhifu/funasr1.0/examples/aishell/conf" \
-    --config-name "train_asr_paraformer_conformer_12e_6d_2048_256.yaml" \
+    python ../../../funasr/bin/compute_audio_cmvn.py \
+    --config-path "${workspace}" \
+    --config-name "${asr_config}" \
     ++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
     ++cmvn_file="${feats_dir}/data/${train_set}/cmvn.json" \
     ++dataset_conf.num_workers=$nj
@@ -116,16 +118,16 @@ fi
 if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
 echo "stage 4: ASR Training"
 
-torchrun \
---nnodes 1 \
---nproc_per_node ${gpu_num} \
-funasr/bin/train.py \
---config-path "/Users/zhifu/funasr1.0/examples/aishell/conf" \
---config-name "train_asr_paraformer_conformer_12e_6d_2048_256.yaml" \
-++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
-++cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
-++token_list="${token_list}" \
-++output_dir="${exp_dir}/exp/${model_dir}"
+  torchrun \
+  --nnodes 1 \
+  --nproc_per_node ${gpu_num} \
+  ../../../funasr/bin/train.py \
+  --config-path "${workspace}" \
+  --config-name "${asr_config}" \
+  ++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
+  ++cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
+  ++token_list="${token_list}" \
+  ++output_dir="${exp_dir}/exp/${model_dir}"
 fi
 
 #

examples/industrial_data_pretraining/paraformer/finetune.sh

@@ -6,10 +6,12 @@
 #git clone https://www.modelscope.cn/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch.git ${local_path}
 
 ## generate jsonl from wav.scp and text.txt
-#python funasr/datasets/audio_datasets/scp2jsonl.py \
-#++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \
-#++data_type_list='["source", "target"]' \
-#++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl
+python funasr/datasets/audio_datasets/scp2jsonl.py \
+++scp_file_list='["/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"]' \
+++data_type_list='["source", "target"]' \
+++jsonl_file_out=/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl
+
+
 # torchrun \
 # --nnodes 1 \
 # --nproc_per_node 1 \

funasr/auto/auto_frontend.py

@@ -19,7 +19,6 @@ from funasr.train_utils.load_pretrained_model import load_pretrained_model
 from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
 from funasr.utils.timestamp_tools import timestamp_sentence
 from funasr.models.campplus.utils import sv_chunk, postprocess, distribute_spk
-from funasr.models.campplus.cluster_backend import ClusterBackend
 from funasr.auto.auto_model import prepare_data_iterator
 
 

funasr/auto/auto_model.py

@@ -20,7 +20,10 @@ from funasr.train_utils.load_pretrained_model import load_pretrained_model
 from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
 from funasr.utils.timestamp_tools import timestamp_sentence
 from funasr.models.campplus.utils import sv_chunk, postprocess, distribute_spk
-from funasr.models.campplus.cluster_backend import ClusterBackend
+try:
+    from funasr.models.campplus.cluster_backend import ClusterBackend
+except:
+    print("If you want to use the speaker diarization, please `pip install hdbscan`")
 
 
 def prepare_data_iterator(data_in, input_len=None, data_type=None, key=None):

funasr/datasets/audio_datasets/scp2jsonl.py

@@ -19,7 +19,7 @@ def gen_jsonl_from_wav_text_list(path, data_type_list=("source", "target"), json
         world_size = 1
 
     cpu_cores = os.cpu_count() or 1
-    
+    print(f"convert wav.scp text to jsonl, ncpu: {cpu_cores}")
     if rank == 0:
         json_dict = {}
         for data_type, data_file in zip(data_type_list, path):
@@ -65,7 +65,7 @@ def parse_context_length(data_list: list, data_type: str):
             sample_num = len(waveform)
             context_len = int(sample_num//16000*1000/10)
         else:
-            context_len = len(line)
+            context_len = len(line.split()) if " " in line else len(line)
         res[key] = {data_type: line, f"{data_type}_len": context_len}
     return res
     
@@ -83,6 +83,8 @@ def main_hydra(cfg: DictConfig):
     kwargs = OmegaConf.to_container(cfg, resolve=True)
 
     scp_file_list = kwargs.get("scp_file_list", ("/Users/zhifu/funasr1.0/test_local/wav.scp", "/Users/zhifu/funasr1.0/test_local/text.txt"))
+    if isinstance(scp_file_list, str):
+        scp_file_list = eval(scp_file_list)
     data_type_list = kwargs.get("data_type_list", ("source", "target"))
     jsonl_file_out = kwargs.get("jsonl_file_out", "/Users/zhifu/funasr1.0/test_local/audio_datasets.jsonl")
     gen_jsonl_from_wav_text_list(scp_file_list, data_type_list=data_type_list, jsonl_file_out=jsonl_file_out)

setup.py

@@ -37,7 +37,6 @@ requirements = {
         # "textgrid",
         # "protobuf",
         "tqdm",
-        "hdbscan",
         "umap_learn",
         "jaconv",
         "hydra-core>=1.3.2",