Merge branch 'main' of https://github.com/alibaba/FunASR into main

2025-09-15 14:48:36 +08:00 · 2023-05-22 19:12:17 +08:00 · 2023-05-22 19:12:17 +08:00 · 1ee853bff5
commit 1ee853bff5
parent 42e1259075 d16e9d8248
14 changed files with 182 additions and 18 deletions
--- a/egs_modelscope/asr/conformer/speech_conformer_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch/demo.py
+++ b/egs_modelscope/asr/conformer/speech_conformer_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch/demo.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model="damo/speech_conformer_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch",
--- a/egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch/demo.py
+++ b/egs_modelscope/asr/paraformer/speech_paraformer_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch/demo.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model="damo/speech_paraformer_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch",
--- a/egs_modelscope/asr/paraformerbert/speech_paraformerbert_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch/infer.py
+++ b/egs_modelscope/asr/paraformerbert/speech_paraformerbert_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch/infer.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model="damo/speech_paraformerbert_asr_nat-zh-cn-16k-aishell1-vocab4234-pytorch",
--- a/egs_modelscope/asr/uniasr/speech_UniASR_asr_2pass-cn-dialect-16k-vocab8358-tensorflow1-online/infer.py
+++ b/egs_modelscope/asr/uniasr/speech_UniASR_asr_2pass-cn-dialect-16k-vocab8358-tensorflow1-online/infer.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model="damo/speech_UniASR_asr_2pass-cn-dialect-16k-vocab8358-tensorflow1-online",
--- a/egs_modelscope/asr/uniasr/speech_UniASR_asr_2pass-zh-cn-16k-common-vocab8358-tensorflow1-online/infer.py
+++ b/egs_modelscope/asr/uniasr/speech_UniASR_asr_2pass-zh-cn-16k-common-vocab8358-tensorflow1-online/infer.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model="damo/speech_UniASR_asr_2pass-zh-cn-16k-common-vocab8358-tensorflow1-online",
--- a/egs_modelscope/asr/uniasr/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab8358-tensorflow1-online/infer.py
+++ b/egs_modelscope/asr/uniasr/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab8358-tensorflow1-online/infer.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model="damo/speech_UniASR_asr_2pass-zh-cn-8k-common-vocab8358-tensorflow1-online",
--- a/egs_modelscope/asr_vad_punc/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/demo.py
+++ b/egs_modelscope/asr_vad_punc/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/demo.py
@ -3,15 +3,14 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/vad_example.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.auto_speech_recognition,
        model='damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch',
        vad_model='damo/speech_fsmn_vad_zh-cn-16k-common-pytorch',
        punc_model='damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch',
        output_dir=output_dir,
-        batch_size=64,
    )
-    rec_result = inference_pipeline(audio_in=audio_in)
+    rec_result = inference_pipeline(audio_in=audio_in, batch_size_token=5000)
    print(rec_result)

--- a/egs_modelscope/vad/speech_fsmn_vad_zh-cn-16k-common/demo.py
+++ b/egs_modelscope/vad/speech_fsmn_vad_zh-cn-16k-common/demo.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/vad_example.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.voice_activity_detection,
        model="damo/speech_fsmn_vad_zh-cn-16k-common-pytorch",
--- a/egs_modelscope/vad/speech_fsmn_vad_zh-cn-16k-common/demo_online.py
+++ b/egs_modelscope/vad/speech_fsmn_vad_zh-cn-16k-common/demo_online.py
@ -7,7 +7,7 @@ logger.setLevel(logging.CRITICAL)
 import soundfile

 if __name__ == '__main__':
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.voice_activity_detection,
        model="damo/speech_fsmn_vad_zh-cn-16k-common-pytorch",
--- a/egs_modelscope/vad/speech_fsmn_vad_zh-cn-8k-common/demo.py
+++ b/egs_modelscope/vad/speech_fsmn_vad_zh-cn-8k-common/demo.py
@ -3,7 +3,7 @@ from modelscope.utils.constant import Tasks

 if __name__ == '__main__':
    audio_in = 'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/vad_example_8k.wav'
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.voice_activity_detection,
        model="damo/speech_fsmn_vad_zh-cn-8k-common",
--- a/egs_modelscope/vad/speech_fsmn_vad_zh-cn-8k-common/demo_online.py
+++ b/egs_modelscope/vad/speech_fsmn_vad_zh-cn-8k-common/demo_online.py
@ -7,7 +7,7 @@ logger.setLevel(logging.CRITICAL)
 import soundfile

 if __name__ == '__main__':
-    output_dir = None
+    output_dir = "./results"
    inference_pipeline = pipeline(
        task=Tasks.voice_activity_detection,
        model="damo/speech_fsmn_vad_zh-cn-8k-common",
--- a/funasr/bin/asr_inference_launch.py
+++ b/funasr/bin/asr_inference_launch.py
@ -600,6 +600,9 @@ def inference_paraformer_vad_punc(
        if 'hotword' in kwargs:
            hotword_list_or_file = kwargs['hotword']
        
+        batch_size_token = kwargs.get("batch_size_token", 6000)
+        print("batch_size_token: ", batch_size_token)
+        
        if speech2text.hotword_list is None:
            speech2text.hotword_list = speech2text.generate_hotwords_list(hotword_list_or_file)
        
@ -642,8 +645,10 @@ def inference_paraformer_vad_punc(
            assert all(isinstance(s, str) for s in keys), keys
            _bs = len(next(iter(batch.values())))
            assert len(keys) == _bs, f"{len(keys)} != {_bs}"
-            
+            beg_vad = time.time()
            vad_results = speech2vadsegment(**batch)
+            end_vad = time.time()
+            print("time cost vad: ", end_vad-beg_vad)
            _, vadsegments = vad_results[0], vad_results[1][0]
            
            speech, speech_lengths = batch["speech"], batch["speech_lengths"]
@ -652,17 +657,29 @@ def inference_paraformer_vad_punc(
            data_with_index = [(vadsegments[i], i) for i in range(n)]
            sorted_data = sorted(data_with_index, key=lambda x: x[0][1] - x[0][0])
            results_sorted = []
-            for j, beg_idx in enumerate(range(0, n, batch_size)):
-                end_idx = min(n, beg_idx + batch_size)
+            batch_size_token_ms = batch_size_token*60
+            batch_size_token_ms_cum = 0
+            beg_idx = 0
+            for j, _ in enumerate(range(0, n)):
+                batch_size_token_ms_cum += (sorted_data[j][0][1] - sorted_data[j][0][0])
+                if j < n-1 and (batch_size_token_ms_cum + sorted_data[j+1][0][1] - sorted_data[j+1][0][0])<batch_size_token_ms:
+                    continue
+                batch_size_token_ms_cum = 0
+                end_idx = j + 1
                speech_j, speech_lengths_j = slice_padding_fbank(speech, speech_lengths, sorted_data[beg_idx:end_idx])
-                
+                beg_idx = end_idx
                batch = {"speech": speech_j, "speech_lengths": speech_lengths_j}
                batch = to_device(batch, device=device)
+                print("batch: ", speech_j.shape[0])
+                beg_asr = time.time()
                results = speech2text(**batch)
+                end_asr = time.time()
+                print("time cost asr: ", end_asr - beg_asr)
                
                if len(results) < 1:
                    results = [["", [], [], [], [], [], []]]
                results_sorted.extend(results)
+            
            restored_data = [0] * n
            for j in range(n):
                index = sorted_data[j][1]
@ -701,7 +718,10 @@ def inference_paraformer_vad_punc(
            text_postprocessed_punc = text_postprocessed
            punc_id_list = []
            if len(word_lists) > 0 and text2punc is not None:
+                beg_punc = time.time()
                text_postprocessed_punc, punc_id_list = text2punc(word_lists, 20)
+                end_punc = time.time()
+                print("time cost punc: ", end_punc-beg_punc)
            
            item = {'key': key, 'value': text_postprocessed_punc}
            if text_postprocessed != "":
--- a/funasr/bin/build_trainer.py
+++ b/funasr/bin/build_trainer.py
@ -0,0 +1,145 @@
+import os
+
+import yaml
+
+
+def update_dct(fin_configs, root):
+    if root == {}:
+        return {}
+    for root_key, root_value in root.items():
+        if not isinstance(root[root_key], dict):
+            fin_configs[root_key] = root[root_key]
+        else:
+            if root_key in fin_configs.keys():
+                result = update_dct(fin_configs[root_key], root[root_key])
+                fin_configs[root_key] = result
+            else:
+                fin_configs[root_key] = root[root_key]
+    return fin_configs
+
+
+def parse_args(mode):
+    if mode == "asr":
+        from funasr.tasks.asr import ASRTask as ASRTask
+    elif mode == "paraformer":
+        from funasr.tasks.asr import ASRTaskParaformer as ASRTask
+    elif mode == "paraformer_vad_punc":
+        from funasr.tasks.asr import ASRTaskParaformer as ASRTask
+    elif mode == "uniasr":
+        from funasr.tasks.asr import ASRTaskUniASR as ASRTask
+    elif mode == "mfcca":
+        from funasr.tasks.asr import ASRTaskMFCCA as ASRTask
+    elif mode == "tp":
+        from funasr.tasks.asr import ASRTaskAligner as ASRTask
+    else:
+        raise ValueError("Unknown mode: {}".format(mode))
+    parser = ASRTask.get_parser()
+    args = parser.parse_args()
+    return args, ASRTask
+
+
+def build_trainer(modelscope_dict,
+                  data_dir,
+                  output_dir,
+                  train_set="train",
+                  dev_set="validation",
+                  distributed=False,
+                  dataset_type="small",
+                  batch_bins=None,
+                  max_epoch=None,
+                  optim=None,
+                  lr=None,
+                  scheduler=None,
+                  scheduler_conf=None,
+                  specaug=None,
+                  specaug_conf=None,
+                  param_dict=None,
+                  **kwargs):
+    mode = modelscope_dict['mode']
+    args, ASRTask = parse_args(mode=mode)
+    # ddp related
+    if args.local_rank is not None:
+        distributed = True
+    else:
+        distributed = False
+    args.local_rank = args.local_rank if args.local_rank is not None else 0
+    local_rank = args.local_rank
+    if "CUDA_VISIBLE_DEVICES" in os.environ.keys():
+        gpu_list = os.environ['CUDA_VISIBLE_DEVICES'].split(",")
+        os.environ['CUDA_VISIBLE_DEVICES'] = str(gpu_list[args.local_rank])
+    else:
+        os.environ['CUDA_VISIBLE_DEVICES'] = str(args.local_rank)
+
+    config = modelscope_dict['am_model_config']
+    finetune_config = modelscope_dict['finetune_config']
+    init_param = modelscope_dict['init_model']
+    cmvn_file = modelscope_dict['cmvn_file']
+    seg_dict_file = modelscope_dict['seg_dict']
+
+    # overwrite parameters
+    with open(config) as f:
+        configs = yaml.safe_load(f)
+    with open(finetune_config) as f:
+        finetune_configs = yaml.safe_load(f)
+        # set data_types
+        if dataset_type == "large":
+            finetune_configs["dataset_conf"]["data_types"] = "sound,text"
+    finetune_configs = update_dct(configs, finetune_configs)
+    for key, value in finetune_configs.items():
+        if hasattr(args, key):
+            setattr(args, key, value)
+
+    # prepare data
+    args.dataset_type = dataset_type
+    if args.dataset_type == "small":
+        args.train_data_path_and_name_and_type = [["{}/{}/wav.scp".format(data_dir, train_set), "speech", "sound"],
+                                                  ["{}/{}/text".format(data_dir, train_set), "text", "text"]]
+        args.valid_data_path_and_name_and_type = [["{}/{}/wav.scp".format(data_dir, dev_set), "speech", "sound"],
+                                                  ["{}/{}/text".format(data_dir, dev_set), "text", "text"]]
+    elif args.dataset_type == "large":
+        args.train_data_file = None
+        args.valid_data_file = None
+    else:
+        raise ValueError(f"Not supported dataset_type={args.dataset_type}")
+    args.init_param = [init_param]
+    args.cmvn_file = cmvn_file
+    if os.path.exists(seg_dict_file):
+        args.seg_dict_file = seg_dict_file
+    else:
+        args.seg_dict_file = None
+    args.data_dir = data_dir
+    args.train_set = train_set
+    args.dev_set = dev_set
+    args.output_dir = output_dir
+    args.gpu_id = args.local_rank
+    args.config = finetune_config
+    if optim is not None:
+        args.optim = optim
+    if lr is not None:
+        args.optim_conf["lr"] = lr
+    if scheduler is not None:
+        args.scheduler = scheduler
+    if scheduler_conf is not None:
+        args.scheduler_conf = scheduler_conf
+    if specaug is not None:
+        args.specaug = specaug
+    if specaug_conf is not None:
+        args.specaug_conf = specaug_conf
+    if max_epoch is not None:
+        args.max_epoch = max_epoch
+    if batch_bins is not None:
+        if args.dataset_type == "small":
+            args.batch_bins = batch_bins
+        elif args.dataset_type == "large":
+            args.dataset_conf["batch_conf"]["batch_size"] = batch_bins
+        else:
+            raise ValueError(f"Not supported dataset_type={args.dataset_type}")
+    if args.normalize in ["null", "none", "None"]:
+        args.normalize = None
+    if args.patience in ["null", "none", "None"]:
+        args.patience = None
+    args.local_rank = local_rank
+    args.distributed = distributed
+    ASRTask.finetune_args = args
+
+    return ASRTask
--- a/funasr/version.txt
+++ b/funasr/version.txt
@ -1 +1 @@
-0.5.4
+0.5.5
 @ -1 +1 @@
 .5.4
 .5.5