Add Python API support for Offline LM rescoring (#1033)

sherpa-onnx/python/sherpa_onnx/offline_recognizer.py

@@ -6,6 +6,7 @@ from typing import List, Optional
 from _sherpa_onnx import (
     FeatureExtractorConfig,
     OfflineCtcFstDecoderConfig,
+    OfflineLMConfig,
     OfflineModelConfig,
     OfflineNemoEncDecCtcModelConfig,
     OfflineParaformerModelConfig,
@@ -56,6 +57,8 @@ class OfflineRecognizer(object):
         model_type: str = "transducer",
         rule_fsts: str = "",
         rule_fars: str = "",
+        lm: str = "",
+        lm_scale: float = 0.1,
     ):
         """
         Please refer to
@@ -143,9 +146,21 @@ class OfflineRecognizer(object):
                 f"--hotwords-file. Currently given: {decoding_method}"
             )
 
+        if lm and decoding_method != "modified_beam_search":
+            raise ValueError(
+                "Please use --decoding-method=modified_beam_search when using "
+                f"--lm. Currently given: {decoding_method}"
+            )
+
+        lm_config = OfflineLMConfig(
+            model=lm,
+            scale=lm_scale,
+        )
+
         recognizer_config = OfflineRecognizerConfig(
             feat_config=feat_config,
             model_config=model_config,
+            lm_config=lm_config,
             decoding_method=decoding_method,
             max_active_paths=max_active_paths,
             hotwords_file=hotwords_file,