add arabic models

examples.py

@@ -197,4 +197,23 @@ examples = [
         4,
         "./test_wavs/tibetan/a_0_cacm-A70_31118.wav",
     ],
+    # arabic
+    [
+        "Arabic",
+        "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06" "greedy_search",
+        4,
+        "./test_wavs/arabic/94D37D38-B203-4FC0-9F3A-538F5C174920_spk-0001_seg-0051454:0052244.wav",
+    ],
+    [
+        "Arabic",
+        "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06" "greedy_search",
+        4,
+        "./test_wavs/arabic/94D37D38-B203-4FC0-9F3A-538F5C174920_spk-0001_seg-0052244:0053004.wav",
+    ],
+    [
+        "Arabic",
+        "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06" "greedy_search",
+        4,
+        "./test_wavs/arabic/94D37D38-B203-4FC0-9F3A-538F5C174920_spk-0001_seg-0053813:0054281.wav",
+    ],
 ]

model.py

@@ -51,6 +51,10 @@ def get_pretrained_model(
         return tibetan_models[repo_id](
             repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
         )
+    elif repo_id in arabic_models:
+        return arabic_models[repo_id](
+            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
+        )
     else:
         raise ValueError(f"Unsupported repo_id: {repo_id}")
 
@@ -402,6 +406,42 @@ def _get_tibetan_pre_trained_model(
     return recognizer
 
 
+@lru_cache(maxsize=10)
+def _get_arabic_pre_trained_model(
+    repo_id: str,
+    decoding_method: str,
+    num_active_paths: int,
+):
+    assert repo_id in [
+        "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06",
+    ], repo_id
+
+    nn_model = _get_nn_model_filename(
+        repo_id=repo_id,
+        filename=filename,
+    )
+
+    tokens = _get_token_filename(repo_id=repo_id, subfolder="data/lang_bpe_5000")
+
+    feat_config = sherpa.FeatureConfig()
+    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
+    feat_config.fbank_opts.mel_opts.num_bins = 80
+    feat_config.fbank_opts.frame_opts.dither = 0
+
+    config = sherpa.OfflineRecognizerConfig(
+        nn_model=nn_model,
+        tokens=tokens,
+        use_gpu=False,
+        feat_config=feat_config,
+        decoding_method=decoding_method,
+        num_active_paths=num_active_paths,
+    )
+
+    recognizer = sherpa.OfflineRecognizer(config)
+
+    return recognizer
+
+
 chinese_models = {
     "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2": _get_wenetspeech_pre_trained_model,  # noqa
     "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12": _get_aishell2_pretrained_model,  # noqa
@@ -427,11 +467,16 @@ tibetan_models = {
     "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29": _get_tibetan_pre_trained_model,  # noqa
 }
 
+arabic_models = {
+    "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06": _get_arabic_pre_trained_model,  # noqa
+}
+
 all_models = {
     **chinese_models,
     **english_models,
     **chinese_english_mixed_models,
     **tibetan_models,
+    **arabic_models,
 }
 
 language_to_models = {
@@ -439,4 +484,5 @@ language_to_models = {
     "English": list(english_models.keys()),
     "Chinese+English": list(chinese_english_mixed_models.keys()),
     "Tibetan": list(tibetan_models.keys()),
+    "Arabic": list(arabic_models.keys()),
 }

test_wavs/arabic/trans.txt

@@ -0,0 +1,3 @@
+94D37D38-B203-4FC0-9F3A-538F5C174920_spk-0001_seg-0053813:0054281 بعد أن عجز وبدأ يصدر مشكلات شعبه ومشكلات مصر
+94D37D38-B203-4FC0-9F3A-538F5C174920_spk-0001_seg-0051454:0052244 وهؤلاء أولياء الشيطان ها هو ذا أحدهم الآن ضيفا عليكم على قناة الجزيرة ولا يستحي في ذلك
+94D37D38-B203-4FC0-9F3A-538F5C174920_spk-0001_seg-0052244:0053004 عندما استغاث الليبيون بالعالم استغاثوا لرفع الظلم وليس لقهر إرادة الأمة ومصادرة الحياة الدستورية