add a cantonese model

model.py

@@ -172,6 +172,10 @@ def get_pretrained_model(
         return chinese_cantonese_english_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
         )
+    elif repo_id in cantonese_models:
+        return cantonese_models[repo_id](
+            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
+        )
     elif repo_id in tibetan_models:
         return tibetan_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
@@ -277,6 +281,46 @@ def _get_aishell2_pretrained_model(
     return recognizer
 
 
+@lru_cache(maxsize=10)
+def _get_zrjin_cantonese_pre_trained_model(
+    repo_id: str, decoding_method: str, num_active_paths: int
+) -> sherpa_onnx.OfflineRecognizer:
+    assert repo_id in ("zrjin/icefall-asr-mdcc-zipformer-2024-03-11",), repo_id
+
+    encoder_model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename="encoder-epoch-45-avg-35.int8.onnx",
+        subfolder="exp",
+    )
+
+    decoder_model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename="decoder-epoch-45-avg-35.onnx",
+        subfolder="exp",
+    )
+
+    joiner_model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename="joiner-epoch-45-avg-35.int8.onnx",
+        subfolder="exp",
+    )
+
+    tokens = _get_token_filename(repo_id=repo_id, subfolder="data/lang_char")
+
+    recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
+        tokens=tokens,
+        encoder=encoder_model,
+        decoder=decoder_model,
+        joiner=joiner_model,
+        num_threads=2,
+        sample_rate=16000,
+        feature_dim=80,
+        decoding_method=decoding_method,
+    )
+
+    return recognizer
+
+
 @lru_cache(maxsize=10)
 def _get_russian_pre_trained_model(
     repo_id: str, decoding_method: str, num_active_paths: int
@@ -1256,12 +1300,17 @@ chinese_cantonese_english_models = {
     "csukuangfj/sherpa-onnx-streaming-paraformer-trilingual-zh-cantonese-en": _get_streaming_paraformer_zh_yue_en_pre_trained_model,
 }
 
+cantonese_models = {
+    "zrjin/icefall-asr-mdcc-zipformer-2024-03-11": _get_zrjin_cantonese_pre_trained_model,
+}
+
 
 all_models = {
     **chinese_models,
     **english_models,
     **chinese_english_mixed_models,
     **chinese_cantonese_english_models,
+    **cantonese_models,
     #  **japanese_models,
     **tibetan_models,
     **arabic_models,
@@ -1275,6 +1324,7 @@ language_to_models = {
     "English": list(english_models.keys()),
     "Chinese+English": list(chinese_english_mixed_models.keys()),
     "Chinese+English+Cantonese": list(chinese_cantonese_english_models.keys()),
+    "Cantonese": list(cantonese_models.keys()),
     #  "Japanese": list(japanese_models.keys()),
     "Tibetan": list(tibetan_models.keys()),
     "Arabic": list(arabic_models.keys()),