Make NeMoTranscription Pipeline Hyperparameters Configurable

src/openbench/pipeline/transcription/transcription_nemo.py

@@ -46,6 +46,22 @@ class NeMoTranscriptionPipelineConfig(TranscriptionConfig):
         default=0.6,
         description="Weight of CTC tokens to prevent false accept errors",
     )
+    keyword_threshold: float = Field(
+        default=-5.0,
+        description="Threshold for keyword detection score",
+    )
+    blank_threshold: float = Field(
+        default=0.8,
+        description="Threshold for blank token probability",
+    )
+    non_blank_threshold: float = Field(
+        default=0.001,
+        description="Threshold for non-blank token probability",
+    )
+    intersection_threshold: float = Field(
+        default=30.0,
+        description="Threshold for intersection between spotted word and word from alignment (in percentage)",
+    )
     spelling_separator: str = Field(
         default="_",
         description="Separator between word and its spellings",
@@ -142,6 +158,9 @@ def _transcribe_with_context_biasing(self, audio_path: Path) -> str:
             beam_threshold=self.config.beam_threshold,
             cb_weight=self.config.context_score,
             ctc_ali_token_weight=self.config.ctc_ali_token_weight,
+            keyword_threshold=self.config.keyword_threshold,
+            blank_threshold=self.config.blank_threshold,
+            non_blank_threshold=self.config.non_blank_threshold,
         )
 
         if not ws_results:
@@ -155,6 +174,7 @@ def _transcribe_with_context_biasing(self, audio_path: Path) -> str:
             self.asr_model,
             ws_results,
             decoder_type=self.config.decoder_type,
+            intersection_threshold=self.config.intersection_threshold,
             blank_idx=self.blank_idx,
             print_stats=False,
         )