Merge pull request #978 from alibaba-damo-academy/dev_lhn

Dev lhn

egs_modelscope/asr/TEMPLATE/README.md

@@ -27,15 +27,18 @@ print(rec_result)
 inference_pipeline = pipeline(
     task=Tasks.auto_speech_recognition,
     model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
-    model_revision='v1.0.6',
+    model_revision='v1.0.7',
     update_model=False,
     mode='paraformer_streaming'
     )
 import soundfile
 speech, sample_rate = soundfile.read("example/asr_example.wav")
 
-chunk_size = [5, 10, 5] #[5, 10, 5] 600ms, [8, 8, 4] 480ms
-param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size}
+chunk_size = [0, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms
+encoder_chunk_look_back = 4 #number of chunks to lookback for encoder self-attention
+decoder_chunk_look_back = 1 #number of encoder chunks to lookback for decoder cross-attention
+param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size,
+              "encoder_chunk_look_back": encoder_chunk_look_back, "decoder_chunk_look_back": decoder_chunk_look_back}
 chunk_stride = chunk_size[1] * 960 # 600ms、480ms
 # first chunk, 600ms
 speech_chunk = speech[0:chunk_stride]
@@ -55,7 +58,7 @@ from modelscope.utils.constant import Tasks
 inference_pipeline = pipeline(
     task=Tasks.auto_speech_recognition,
     model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
-    model_revision='v1.0.6',
+    model_revision='v1.0.7',
     update_model=False,
     mode="paraformer_fake_streaming"
 )

egs_modelscope/asr/TEMPLATE/README_zh.md

@@ -27,15 +27,18 @@ print(rec_result)
 inference_pipeline = pipeline(
     task=Tasks.auto_speech_recognition,
     model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
-    model_revision='v1.0.6',
+    model_revision='v1.0.7',
     update_model=False,
     mode='paraformer_streaming'
     )
 import soundfile
 speech, sample_rate = soundfile.read("example/asr_example.wav")
 
-chunk_size = [5, 10, 5] #[5, 10, 5] 600ms, [8, 8, 4] 480ms
-param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size}
+chunk_size = [0, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms
+encoder_chunk_look_back = 4 #number of chunks to lookback for encoder self-attention
+decoder_chunk_look_back = 1 #number of encoder chunks to lookback for decoder cross-attention
+param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size,
+              "encoder_chunk_look_back": encoder_chunk_look_back, "decoder_chunk_look_back": decoder_chunk_look_back}
 chunk_stride = chunk_size[1] * 960 # 600ms、480ms
 # first chunk, 600ms
 speech_chunk = speech[0:chunk_stride]
@@ -55,7 +58,7 @@ from modelscope.utils.constant import Tasks
 inference_pipeline = pipeline(
     task=Tasks.auto_speech_recognition,
     model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
-    model_revision='v1.0.6',
+    model_revision='v1.0.7',
     update_model=False,
     mode="paraformer_fake_streaming"
 )

egs_modelscope/asr/paraformer/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/demo.py

@@ -4,7 +4,7 @@ from modelscope.utils.constant import Tasks
 inference_pipeline = pipeline(
     task=Tasks.auto_speech_recognition,
     model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
-    model_revision='v1.0.6',
+    model_revision='v1.0.7',
     update_model=False,
     mode="paraformer_fake_streaming"
 )

egs_modelscope/asr/paraformer/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/demo_online.py

@@ -14,7 +14,7 @@ os.environ["MODELSCOPE_CACHE"] = "./"
 inference_pipeline = pipeline(
     task=Tasks.auto_speech_recognition,
     model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
-    model_revision='v1.0.6',
+    model_revision='v1.0.7',
     update_model=False,
     mode="paraformer_streaming"
 )

egs_modelscope/asr/paraformer/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/demo_online_v2.py

@@ -24,9 +24,12 @@ speech, sample_rate = soundfile.read(os.path.join(model_dir, "example/asr_exampl
 speech_length = speech.shape[0]
 
 sample_offset = 0
-chunk_size = [0, 8, 4] #[5, 10, 5] 600ms, [8, 8, 4] 480ms
+chunk_size = [0, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms
+encoder_chunk_look_back = 4 #number of chunks to lookback for encoder self-attention
+decoder_chunk_look_back = 1 #number of encoder chunks to lookback for decoder cross-attention
 stride_size =  chunk_size[1] * 960
-param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size, "encoder_chunk_look_back": 4, "decoder_chunk_look_back": 1}
+param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size, 
+              "encoder_chunk_look_back": encoder_chunk_look_back, "decoder_chunk_look_back": decoder_chunk_look_back}
 final_result = ""
 
 for sample_offset in range(0, speech_length, min(stride_size, speech_length - sample_offset)):

egs_modelscope/asr/paraformer/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/finetune.py

@@ -14,7 +14,7 @@ def modelscope_finetune(params):
     ds_dict = MsDataset.load(params.data_path)
     kwargs = dict(
         model=params.model,
-        model_revision='v1.0.6',
+        model_revision='v1.0.7',
         update_model=False,
         data_dir=ds_dict,
         dataset_type=params.dataset_type,

egs_modelscope/asr/paraformer/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online/infer.py

@@ -11,7 +11,7 @@ def modelscope_infer(args):
         model=args.model,
         output_dir=args.output_dir,
         batch_size=args.batch_size,
-        model_revision='v1.0.6',
+        model_revision='v1.0.7',
         update_model=False,
         mode="paraformer_fake_streaming",
         param_dict={"decoding_model": args.decoding_mode, "hotword": args.hotword_txt}