add a new model

model.py

@@ -164,6 +164,10 @@ def get_pretrained_model(
         return chinese_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
         )
+    elif repo_id in chinese_dialect_models:
+        return chinese_dialect_models[repo_id](
+            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
+        )
     elif repo_id in english_models:
         return english_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
@@ -1092,6 +1096,31 @@ def _get_paraformer_en_pre_trained_model(
     return recognizer
 
 
+@lru_cache(maxsize=5)
+def _get_chinese_dialect_models(
+    repo_id: str, decoding_method: str, num_active_paths: int
+) -> sherpa_onnx.OfflineRecognizer:
+    assert repo_id in [
+        "csukuangfj/sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04",
+    ], repo_id
+
+    nn_model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename="model.int8.onnx",
+        subfolder=".",
+    )
+
+    tokens = _get_token_filename(repo_id=repo_id, subfolder=".")
+
+    recognizer = sherpa_onnx.OfflineRecognizer.from_telespeech_ctc(
+        paraformer=nn_model,
+        tokens=tokens,
+        num_threads=2,
+    )
+
+    return recognizer
+
+
 @lru_cache(maxsize=10)
 def _get_paraformer_pre_trained_model(
     repo_id: str,
@@ -1239,6 +1268,10 @@ def _get_multi_zh_hans_pre_trained_model(
     return recognizer
 
 
+chinese_models = {
+    "csukuangfj/sherpa-onnx-telespeech-ctc-int8-zh-2024-06-04": _get_chinese_dialect_models,
+}
+
 chinese_models = {
     "csukuangfj/sherpa-onnx-paraformer-zh-2024-03-09": _get_paraformer_pre_trained_model,
     "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2": _get_wenetspeech_pre_trained_model,  # noqa
@@ -1339,6 +1372,7 @@ all_models = {
 }
 
 language_to_models = {
+    "超多种中文方言": list(chinese_dialect_models.keys()),
     "Chinese": list(chinese_models.keys()),
     "English": list(english_models.keys()),
     "Chinese+English": list(chinese_english_mixed_models.keys()),

requirements.txt

@@ -9,4 +9,7 @@ sentencepiece>=0.1.96
 numpy
 
 huggingface_hub
-sherpa-onnx>=1.9.21
+
+https://huggingface.co/csukuangfj/sherpa-onnx-wheels /resolve/main/sherpa_onnx-1.9.26-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+
+#sherpa-onnx>=1.9.21