Update model.py

model.py

@@ -206,6 +206,10 @@ def get_pretrained_model(
         return cantonese_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
         )
+    elif repo_id in revolab_models:
+        return revolab_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
@@ -473,6 +477,116 @@ def _get_yifan_thai_pretrained_model(
 
     return recognizer
 
+@lru_cache(maxsize=10)
+def _get_revolab_pretrained_model(
+    repo_id: str, decoding_method: str, num_active_paths: int
+) -> sherpa_onnx.OfflineRecognizer:
+    assert 'Revolab' in repo_id
+
+    if repo_id == "Revolab/zipformer-large-145M":
+        real_repo = 'Revolab/malaysian-pruned_transducer_stateless7'
+        encoder_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="encoder-epoch-19-avg-1.onnx",
+            subfolder="zipformer-large-20k/export",
+        )
+
+        decoder_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="decoder-epoch-19-avg-1.onnx",
+            subfolder="zipformer-large-20k/export",
+        )
+
+        joiner_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="joiner-epoch-19-avg-1.onnx",
+            subfolder="zipformer-large-20k/export",
+        )
+
+        tokens = _get_token_filename(repo_id=real_repo, subfolder="zipformer-large-20k/exp/lang_bpe_500")
+
+        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
+    elif repo_id == "Revolab/zipformer-large-finetuned-145M":
+        real_repo = 'Revolab/malaysian-pruned_transducer_stateless7'
+
+        encoder_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="encoder-epoch-17-avg-1.onnx",
+            subfolder="zipformer-large-finetune-SFO/export",
+        )
+        decoder_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="decoder-epoch-19-avg-3.onnx",
+            subfolder="zipformer-large-finetune-SFO/export",
+        )
+
+        joiner_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="joiner-epoch-19-avg-3.onnx",
+            subfolder="zipformer-large-finetune-SFO/export",
+        )
+
+        tokens = _get_token_filename(repo_id=real_repo, subfolder="zipformer-large-20k/exp/lang_bpe_500")
+
+        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
+
+    elif repo_id == "Revolab/pruned-transducer-65M":
+        real_repo = 'Revolab/malaysian-pruned_transducer_stateless7'
+
+        encoder_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="encoder-epoch-19-avg-3.onnx",
+            subfolder="PT7-stage1/export",
+        )
+        decoder_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="decoder-epoch-19-avg-3.onnx",
+            subfolder="PT7-stage1/export",
+        )
+
+        joiner_model = _get_nn_model_filename(
+            repo_id=real_repo,
+            filename="joiner-epoch-19-avg-3.onnx",
+            subfolder="PT7-stage1/export",
+        )
+
+        tokens = _get_token_filename(repo_id=real_repo, subfolder="pts-16k-all/exp/lang_bpe_500")
+
+        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_zrjin_cantonese_pre_trained_model(
@@ -2293,6 +2407,14 @@ cantonese_models = {
     "zrjin/icefall-asr-mdcc-zipformer-2024-03-11": _get_zrjin_cantonese_pre_trained_model,
 }
 
+revolab_models = {
+    "Revolab/zipformer-large-145M": _get_revolab_pretrained_model,
+    "Revolab/pruned-transducer-65M": _get_revolab_pretrained_model,
+    "Revolab/zipformer-large-finetuned-145M":_get_revolab_pretrained_model,
+
+}
+
+
 korean_models = {
     "k2-fsa/sherpa-onnx-zipformer-korean-2024-06-24": _get_offline_pre_trained_model,
     "k2-fsa/sherpa-onnx-streaming-zipformer-korean-2024-06-16": _get_streaming_zipformer_pre_trained_model,
@@ -2325,6 +2447,7 @@ all_models = {
     **chinese_cantonese_english_models,
     **chinese_cantonese_english_japanese_korean_models,
     **cantonese_models,
+    **revolab_models,
     **japanese_models,
     **tibetan_models,
     **arabic_models,
@@ -2351,6 +2474,7 @@ language_to_models = {
     ),
     "Arabic": list(arabic_models.keys()),
     "Cantonese": list(cantonese_models.keys()),
+    "Malay": list(revolab_models.keys()),
     "French": list(french_models.keys()),
     "German": list(german_models.keys()),
     "Japanese": list(japanese_models.keys()),