Add small changes before adding the sc_args functionality

Changes include:
- Changing nargs for hypfile and reffile args
- Small rewording of comments and help messages
- Removing skip_normalization as it was redundant
- Changing the way interactive behaves (have it be a True value in the code when used)
- Change normalization of numbers slightly (add a space after duizend, this is how it's done for Dutch)
- Reorder sclite and variation related LOCs
- Update README with new arguments added

ASR_NL_benchmark/__main__.py

@@ -6,44 +6,42 @@
 if __name__ == "__main__":
     # Set parser
     parser = argparse.ArgumentParser(description='normalize ref and hyp file')
-    parser.add_argument('-hyp', '--hypfile', nargs='+',
+    parser.add_argument('-hyp', '--hypfile', nargs=2,
                         metavar=('hypfile_name', 'extension'),
-                        default=['ASR_NL_benchmark/data/test_hyp.ctm', 'ctm'], help='help: path to the hypothesis file and its extension')
-    parser.add_argument('-ref', '--reffile', nargs='+',
+                        default=['ASR_NL_benchmark/data/test_hyp.ctm', 'ctm'],
+                        help='path to the hypothesis file and its extension')
+    parser.add_argument('-ref', '--reffile', nargs=2,
                         metavar=('reffile_name', 'extension'),
                         default=['ASR_NL_benchmark/data/test_ref.stm', 'stm'],
-                        help='help: path to the reference file and its extension')
+                        help='path to the reference file and its extension')
     parser.add_argument('-kind', '--kind',
                         metavar=('speechrecognizer'),
                         default='',
-                        help='help: enter the name of your speech recognizer')
+                        help='enter the name of your speech recognizer')
     parser.add_argument('-interactive',
-                        metavar='value',
-                        default='',
-                        help='help: True if you want to use the GUI')
+                        action = 'store_true',
+                        help='if you want to use the GUI')
     parser.add_argument('-skip_ref_normalization', 
                         action = 'store_true',
-                        help = 'Skip the normalization step for the reference file')
+                        help = 'skip the normalization step for the reference file')
     parser.add_argument('-skip_hyp_normalization', 
                         action = 'store_true',
-                        help = 'Skip the normalization step for the hypothesis file')
-    parser.add_argument('-skip-normalization',
-                        action = 'store_true',
-                        help = 'Skip the normalization step for both hypothesis and reference files')
+                        help = 'skip the normalization step for the hypothesis file')
+    parser.add_argument('-sc_args', nargs='*',
+                        default=[],
+                        help='extra sclite arguments you want to use (without the -)')
 
     args = parser.parse_args()
 
-    if bool(args.interactive):
+    if args.interactive:
         print('Opening interface')
         interface.main()
     else:
         print('Running benchmarking')
-        skip_ref_norm = args.skip_ref_normalization
-        skip_hyp_norm = args.skip_hyp_normalization
-        if args.skip_normalization:
-            skip_ref_norm = args.skip_ref_normalization
-            skip_hyp_norm = args.skip_hyp_normalization
-        benchmarking = pipeline.Pipeline(args.hypfile[0], args.hypfile[1], args.reffile[0], args.reffile[1], kind=args.kind, skip_ref_norm=skip_ref_norm, skip_hyp_norm=skip_hyp_norm)
+        benchmarking = pipeline.Pipeline(args.hypfile[0], args.hypfile[1], \
+                                         args.reffile[0], args.reffile[1], \
+                                         kind=args.kind, \
+                                         skip_ref_norm=args.skip_ref_normalization, skip_hyp_norm=args.skip_hyp_normalization)
         benchmarking.main()
         pipeline.process_results(kind=args.kind)
 

ASR_NL_benchmark/normalize.py

@@ -20,6 +20,7 @@ def check_and_covert_interger(word):
             float(word)
             new_word = num2words(word, to='cardinal', lang='nl')
             new_word = new_word.replace('komma', 'punt')
+            new_word = new_word.replace('duizend', 'duizend ')
             logging.info(f'converted the number {word} to {new_word}')
             return new_word
         except:
@@ -29,6 +30,7 @@ def check_and_covert_interger(word):
         try:
             float(new_word)
             new_word = num2words(new_word, to='cardinal', lang='nl')
+            new_word = new_word.replace('duizend', 'duizend ')
             logging.info(f'converted number {word} to {new_word}')
             return new_word
         except:
@@ -66,6 +68,7 @@ def replace_numbers(text):
         if word.isdigit():
             number_of_numbers += 1
             text_list[position] = num2words(word, to='cardinal', lang='nl')
+            text_list[position] = text_list[position].replace('duizend', 'duizend ')
         elif check_and_covert_interger(word):
             text_list[position] = check_and_covert_interger(word)
     text_without_numbers = " ".join(text_list)

ASR_NL_benchmark/pipeline.py

@@ -47,18 +47,20 @@ def run_pipeline(hypfile, reffile, skip_ref_norm, skip_hyp_norm):
     #Create results folder if not exists:
     if not os.path.exists(os.path.join(os.path.sep,'input','results')):
         os.makedirs(os.path.join(os.path.sep,'input','results'))
+
+    # sclite command to be logged and executed
+    command = f"sclite -D -h {hypfile.variation_path} {hypfile.extension} -r {reffile.variation_path} {reffile.extension} \
+        -m hyp -O {os.path.join(os.path.sep,'input','results')} -o prf dtl spk"
 
     # Run variation scripts
-    logging.info(
-        f"running: sclite -h {hypfile.normalized_path} {hypfile.extension} -r {reffile.normalized_path} {reffile.extension} -m hyp -O {os.path.join(os.path.sep,'input','results')}  -o dtl spk")
-    run = os.system(
-        f"csrfilt.sh -s -i ctm {os.path.join('ASR_NL_benchmark','variations.glm')} < {hypfile.normalized_path} > {hypfile.variation_path}")
-
-    # Run sclite
-    run = os.system(
-        f"csrfilt.sh -s -i stm {os.path.join('ASR_NL_benchmark','variations.glm')} < {reffile.normalized_path} > {reffile.variation_path}")
-    run = os.system(
-        f"sclite -D -h {hypfile.variation_path} {hypfile.extension} -r {reffile.variation_path} {reffile.extension} -m hyp -O {os.path.join(os.path.sep,'input','results')} -o prf dtl spk")
+    # Hypothesis
+    run = os.system(f"csrfilt.sh -s -i ctm {os.path.join('ASR_NL_benchmark','variations.glm')} < {hypfile.normalized_path} > {hypfile.variation_path}")
+    # Reference
+    run = os.system(f"csrfilt.sh -s -i stm {os.path.join('ASR_NL_benchmark','variations.glm')} < {reffile.normalized_path} > {reffile.variation_path}")
+
+    # Log & run sclite
+    logging.info("running:" + command)
+    run = os.system(command)
 
 def calculate_wer(df):
     """ Calculates the word error rate and adds the collumn 'product' to the dataframe

README.md

@@ -61,6 +61,11 @@ The final results are saved in .csv format inside a folder named `results` store
 - .spk files - Report with scoring for a speaker as returned by sclite
 - .csv files - Overall results of the benchmarking as shown in the interface
 
+## Extra arguments
+There are extra arguments that you can add to the command line:
+- `-skip_hyp_normalization`: Skips the normalization step for the hypothesis file(s) (STILL APPLIES VARIATIONS)
+- `-skip_ref_normalization`: Skips the normalization step for the reference file(s) (STILL APPLIES VARIATIONS)
+- `-sc_args`: With this argument, you can add extra sclite-specific flags. For more information, check the [documentation of sclite](https://github.com/usnistgov/SCTK/blob/master/doc/sclite.htm) (to view it properly, we suggest locally downloading the entire `doc` folder of the SCTK repository).
 
 ## More about the pipeline
 ### Normalization