add more russian models

examples.py

@@ -82,7 +82,7 @@ examples = [
     ],
     [
         "Russian",
-        "alphacep/vosk-model-ru",
+        "csukuangfj/sherpa-onnx-nemo-transducer-giga-am-russian-2024-10-24",
         "greedy_search",
         4,
         "No",

model.py

@@ -430,6 +430,31 @@ def _get_zrjin_cantonese_pre_trained_model(
     return recognizer
 
 
+@lru_cache(maxsize=10)
+def _get_russian_pre_trained_model_ctc(
+    repo_id: str, decoding_method: str, num_active_paths: int
+) -> sherpa_onnx.OfflineRecognizer:
+    assert repo_id in (
+        "csukuangfj/sherpa-onnx-nemo-ctc-giga-am-russian-2024-10-24",
+    ), repo_id
+
+    model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename="model.int8.onnx",
+        subfolder=model_dir,
+    )
+
+    tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
+
+    recognizer = sherpa_onnx.OfflineRecognizer.from_nemo_ctc(
+        model=model,
+        tokens=tokens,
+        num_threads=2,
+    )
+
+    return recognizer
+
+
 @lru_cache(maxsize=10)
 def _get_russian_pre_trained_model(
     repo_id: str, decoding_method: str, num_active_paths: int
@@ -437,16 +462,25 @@ def _get_russian_pre_trained_model(
     assert repo_id in (
         "alphacep/vosk-model-ru",
         "alphacep/vosk-model-small-ru",
+        "csukuangfj/sherpa-onnx-nemo-transducer-giga-am-russian-2024-10-24",
     ), repo_id
 
     if repo_id == "alphacep/vosk-model-ru":
         model_dir = "am-onnx"
+        encoder = "encoder.onnx"
+        model_type = "transducer"
     elif repo_id == "alphacep/vosk-model-small-ru":
         model_dir = "am"
+        encoder = "encoder.onnx"
+        model_type = "transducer"
+    elif repo_id == "csukuangfj/sherpa-onnx-nemo-transducer-giga-am-russian-2024-10-24":
+        model_dir = "."
+        encoder = "encoder.int8.onnx"
+        model_type = "nemo_transducer"
 
     encoder_model = _get_nn_model_filename(
         repo_id=repo_id,
-        filename="encoder.onnx",
+        filename=encoder,
         subfolder=model_dir,
     )
 
@@ -462,7 +496,10 @@ def _get_russian_pre_trained_model(
         subfolder=model_dir,
     )
 
-    tokens = _get_token_filename(repo_id=repo_id, subfolder="lang")
+    if repo_id == "csukuangfj/sherpa-onnx-nemo-transducer-giga-am-russian-2024-10-24":
+        tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
+    else:
+        tokens = _get_token_filename(repo_id=repo_id, subfolder="lang")
 
     recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
         tokens=tokens,
@@ -473,6 +510,7 @@ def _get_russian_pre_trained_model(
         sample_rate=16000,
         feature_dim=80,
         decoding_method=decoding_method,
+        model_type=model_type,
     )
 
     return recognizer
@@ -1639,6 +1677,8 @@ japanese_models = {
 }
 
 russian_models = {
+    "csukuangfj/sherpa-onnx-nemo-transducer-giga-am-russian-2024-10-24": _get_russian_pre_trained_model,
+    "csukuangfj/sherpa-onnx-nemo-ctc-giga-am-russian-2024-10-24": _get_russian_pre_trained_model_ctc,
     "alphacep/vosk-model-ru": _get_russian_pre_trained_model,
     "alphacep/vosk-model-small-ru": _get_russian_pre_trained_model,
 }