code update

runtime/python/utils/test_cer.py

@@ -17,8 +17,8 @@ args = parser.parse_args()
 
 from funasr.runtime.python.libtorch.funasr_torch import Paraformer
 if args.backend == "onnx":
-	from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
-	
+    from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
+    
 model = Paraformer(args.model_dir, batch_size=1, quantize=args.quantize, intra_op_num_threads=args.intra_op_num_threads)
 
 wav_file_f = open(args.wav_file, 'r')
@@ -26,23 +26,23 @@ wav_files = wav_file_f.readlines()
 
 output_dir = args.output_dir
 if not os.path.exists(output_dir):
-	os.makedirs(output_dir)
+    os.makedirs(output_dir)
 if os.name == 'nt':   # Windows
-	newline = '\r\n'
+    newline = '\r\n'
 else:   # Linux Mac
-	newline = '\n'
+    newline = '\n'
 text_f = open(os.path.join(output_dir, "text"), "w", newline=newline)
 token_f = open(os.path.join(output_dir, "token"), "w", newline=newline)
 
 for i, wav_path_i in enumerate(wav_files):
-	wav_name, wav_path = wav_path_i.strip().split()
-	result = model(wav_path)
-	text_i = "{} {}\n".format(wav_name, result[0]['preds'][0])
-	token_i = "{} {}\n".format(wav_name, result[0]['preds'][1])
-	text_f.write(text_i)
-	text_f.flush()
-	token_f.write(token_i)
-	token_f.flush()
+    wav_name, wav_path = wav_path_i.strip().split()
+    result = model(wav_path)
+    text_i = "{} {}\n".format(wav_name, result[0]['preds'][0])
+    token_i = "{} {}\n".format(wav_name, result[0]['preds'][1])
+    text_f.write(text_i)
+    text_f.flush()
+    token_f.write(token_i)
+    token_f.flush()
 text_f.close()
 token_f.close()
-	
+    

runtime/python/utils/test_rtf.py

@@ -16,8 +16,8 @@ args = parser.parse_args()
 
 from funasr.runtime.python.libtorch.funasr_torch import Paraformer
 if args.backend == "onnx":
-	from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
-	
+    from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
+    
 model = Paraformer(args.model_dir, batch_size=1, quantize=args.quantize, intra_op_num_threads=args.intra_op_num_threads)
 
 wav_file_f = open(args.wav_file, 'r')
@@ -28,28 +28,28 @@ total = 0.0
 num = 30
 wav_path = wav_files[0].split("\t")[1].strip() if "\t" in wav_files[0] else wav_files[0].split(" ")[1].strip()
 for i in range(num):
-	beg_time = time.time()
-	result = model(wav_path)
-	end_time = time.time()
-	duration = end_time-beg_time
-	total += duration
-	print(result)
-	print("num: {}, time, {}, avg: {}, rtf: {}".format(len(wav_path), duration, total/(i+1), (total/(i+1))/5.53))
+    beg_time = time.time()
+    result = model(wav_path)
+    end_time = time.time()
+    duration = end_time-beg_time
+    total += duration
+    print(result)
+    print("num: {}, time, {}, avg: {}, rtf: {}".format(len(wav_path), duration, total/(i+1), (total/(i+1))/5.53))
 
 # infer time
 beg_time = time.time()
 for i, wav_path_i in enumerate(wav_files):
-	wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
-	result = model(wav_path)
+    wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
+    result = model(wav_path)
 end_time = time.time()
 duration = (end_time-beg_time)*1000
 print("total_time_comput_ms: {}".format(int(duration)))
 
 duration_time = 0.0
 for i, wav_path_i in enumerate(wav_files):
-	wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
-	waveform, _ = librosa.load(wav_path, sr=16000)
-	duration_time += len(waveform)/16.0
+    wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
+    waveform, _ = librosa.load(wav_path, sr=16000)
+    duration_time += len(waveform)/16.0
 print("total_time_wav_ms: {}".format(int(duration_time)))
 
 print("total_rtf: {:.5}".format(duration/duration_time))

runtime/python/utils/test_rtf_gpu.py

@@ -17,8 +17,8 @@ args = parser.parse_args()
 
 from funasr.runtime.python.libtorch.funasr_torch import Paraformer
 if args.backend == "onnx":
-	from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
-	
+    from funasr.runtime.python.onnxruntime.funasr_onnx import Paraformer
+    
 model = Paraformer(args.model_dir, batch_size=args.batch_size, quantize=args.quantize, intra_op_num_threads=args.intra_op_num_threads)
 
 wav_file_f = open(args.wav_file, 'r')
@@ -29,20 +29,20 @@ total = 0.0
 num = 30
 wav_path = wav_files[0].split("\t")[1].strip() if "\t" in wav_files[0] else wav_files[0].split(" ")[1].strip()
 for i in range(num):
-	beg_time = time.time()
-	result = model(wav_path)
-	end_time = time.time()
-	duration = end_time-beg_time
-	total += duration
-	print(result)
-	print("num: {}, time, {}, avg: {}, rtf: {}".format(len(wav_path), duration, total/(i+1), (total/(i+1))/5.53))
+    beg_time = time.time()
+    result = model(wav_path)
+    end_time = time.time()
+    duration = end_time-beg_time
+    total += duration
+    print(result)
+    print("num: {}, time, {}, avg: {}, rtf: {}".format(len(wav_path), duration, total/(i+1), (total/(i+1))/5.53))
 
 # infer time
 wav_path = []
 beg_time = time.time()
 for i, wav_path_i in enumerate(wav_files):
-	wav_path_i = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
-	wav_path += [wav_path_i]
+    wav_path_i = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
+    wav_path += [wav_path_i]
 result = model(wav_path)
 end_time = time.time()
 duration = (end_time-beg_time)*1000
@@ -50,9 +50,9 @@ print("total_time_comput_ms: {}".format(int(duration)))
 
 duration_time = 0.0
 for i, wav_path_i in enumerate(wav_files):
-	wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
-	waveform, _ = librosa.load(wav_path, sr=16000)
-	duration_time += len(waveform)/16.0
+    wav_path = wav_path_i.split("\t")[1].strip() if "\t" in wav_path_i else wav_path_i.split(" ")[1].strip()
+    waveform, _ = librosa.load(wav_path, sr=16000)
+    duration_time += len(waveform)/16.0
 print("total_time_wav_ms: {}".format(int(duration_time)))
 
 print("total_rtf: {:.5}".format(duration/duration_time))