add vietnamese asr

model.py

@@ -228,6 +228,10 @@ def get_pretrained_model(
         return thai_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
         )
+    elif repo_id in vietnamese_models:
+        return vietnamese_models[repo_id](
+            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
+        )
     else:
         raise ValueError(f"Unsupported repo_id: {repo_id}")
 
@@ -352,6 +356,64 @@ def _get_offline_pre_trained_model(
     return recognizer
 
 
+@lru_cache(maxsize=10)
+def _get_vietnamese_pretrained_model(
+    repo_id: str, decoding_method: str, num_active_paths: int
+) -> sherpa_onnx.OfflineRecognizer:
+    assert repo_id in (
+        "csukuangfj/sherpa-onnx-zipformer-vi-int8-2025-04-20",
+        "csukuangfj/sherpa-onnx-zipformer-vi-2025-04-20",
+    ), repo_id
+
+    decoder_model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename="decoder-epoch-12-avg-8.onnx",
+        subfolder=".",
+    )
+
+    if repo_id == "csukuangfj/sherpa-onnx-zipformer-vi-int8-2025-04-20":
+        encoder_model = _get_nn_model_filename(
+            repo_id=repo_id,
+            filename="encoder-epoch-12-avg-8.int8.onnx",
+            subfolder=".",
+        )
+
+        joiner_model = _get_nn_model_filename(
+            repo_id=repo_id,
+            filename="joiner-epoch-12-avg-8.int8.onnx",
+            subfolder=".",
+        )
+    elif repo_id == "csukuangfj/sherpa-onnx-zipformer-vi-2025-04-20":
+        encoder_model = _get_nn_model_filename(
+            repo_id=repo_id,
+            filename="encoder-epoch-12-avg-8.onnx",
+            subfolder=".",
+        )
+
+        joiner_model = _get_nn_model_filename(
+            repo_id=repo_id,
+            filename="joiner-epoch-12-avg-8.onnx",
+            subfolder=".",
+        )
+    else:
+        raise ValueError(f"repo_id: {repo_id}")
+
+    tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
+
+    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_yifan_thai_pretrained_model(
     repo_id: str, decoding_method: str, num_active_paths: int
@@ -1899,6 +1961,11 @@ thai_models = {
     "yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20": _get_yifan_thai_pretrained_model,
 }
 
+vietnamese_models = {
+    "csukuangfj/sherpa-onnx-zipformer-vi-int8-2025-04-20": _get_vietnamese_pretrained_model,
+    "csukuangfj/sherpa-onnx-zipformer-vi-2025-04-20": _get_vietnamese_pretrained_model,
+}
+
 
 all_models = {
     **multi_lingual_models,
@@ -1916,6 +1983,7 @@ all_models = {
     **russian_models,
     **korean_models,
     **thai_models,
+    **vietnamese_models,
 }
 
 language_to_models = {
@@ -1937,4 +2005,5 @@ language_to_models = {
     "Russian": list(russian_models.keys()),
     "Korean": list(korean_models.keys()),
     "Thai": list(thai_models.keys()),
+    "Vietnamese": list(vietnamese_models.keys()),
 }