sync

README.md

@@ -1,8 +1,8 @@
 ---
-title: Test
-emoji: 📚
-colorFrom: green
-colorTo: gray
+title: Automatic Speech Recognition
+emoji: 🌖
+colorFrom: yellow
+colorTo: green
 sdk: gradio
 sdk_version: 3.0.26
 app_file: app.py

examples.py

@@ -176,4 +176,25 @@ examples = [
         4,
         "./test_wavs/tal_csasr/210_36476_210_8341_1_1533271973_7057520_145.wav",
     ],
+    [
+        "Tibetan",
+        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29",
+        "greedy_search",
+        4,
+        "./test_wavs/tibetan/a_0_cacm-A70_31116.wav",
+    ],
+    [
+        "Tibetan",
+        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29",
+        "greedy_search",
+        4,
+        "./test_wavs/tibetan/a_0_cacm-A70_31117.wav",
+    ],
+    [
+        "Tibetan",
+        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29",
+        "greedy_search",
+        4,
+        "./test_wavs/tibetan/a_0_cacm-A70_31118.wav",
+    ],
 ]

model.py

@@ -31,6 +31,8 @@ def get_pretrained_model(repo_id: str) -> OfflineAsr:
         return english_models[repo_id](repo_id)
     elif repo_id in chinese_english_mixed_models:
         return chinese_english_mixed_models[repo_id](repo_id)
+    elif repo_id in tibetan_models:
+        return tibetan_models[repo_id](repo_id)
     else:
         raise ValueError(f"Unsupported repo_id: {repo_id}")
 
@@ -122,12 +124,28 @@ def _get_gigaspeech_pre_trained_model(repo_id: str) -> OfflineAsr:
 @lru_cache(maxsize=10)
 def _get_librispeech_pre_trained_model(repo_id: str) -> OfflineAsr:
     assert repo_id in [
+        "WeijiZhuang/icefall-asr-librispeech-pruned-transducer-stateless8-2022-12-02",  # noqa
         "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13",  # noqa
+        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11",  # noqa
+        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless8-2022-11-14",  # noqa
     ], repo_id
 
+    filename = "cpu_jit.pt"
+    if (
+        repo_id
+        == "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11"
+    ):
+        filename = "cpu_jit-torch-1.10.0.pt"
+
+    if (
+        repo_id
+        == "WeijiZhuang/icefall-asr-librispeech-pruned-transducer-stateless8-2022-12-02"
+    ):
+        filename = "cpu_jit-torch-1.10.pt"
+
     nn_model_filename = _get_nn_model_filename(
         repo_id=repo_id,
-        filename="cpu_jit.pt",
+        filename=filename,
     )
     bpe_model_filename = _get_bpe_model_filename(repo_id=repo_id)
 
@@ -224,16 +242,47 @@ def _get_aidatatang_200zh_pretrained_mode(repo_id: str):
     )
 
 
+@lru_cache(maxsize=10)
+def _get_tibetan_pre_trained_model(repo_id: str):
+    assert repo_id in [
+        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless7-2022-12-02",
+        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29",
+    ], repo_id
+
+    filename = "cpu_jit.pt"
+    if (
+        repo_id
+        == "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29"
+    ):
+        nn_model_filename = _get_nn_model_filename(
+            repo_id=repo_id,
+            filename="cpu_jit-epoch-28-avg-23-torch-1.10.0.pt",
+        )
+
+    bpe_model_filename = _get_bpe_model_filename(repo_id=repo_id)
+
+    return OfflineAsr(
+        nn_model_filename=nn_model_filename,
+        bpe_model_filename=bpe_model_filename,
+        token_filename=None,
+        sample_rate=sample_rate,
+        device="cpu",
+    )
+
+
 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
     "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12": _get_aishell2_pretrained_model,  # noqa
-    "luomingshuang/icefall_asr_alimeeting_pruned_transducer_stateless2": _get_alimeeting_pre_trained_model,  # noqa
     "luomingshuang/icefall_asr_aidatatang-200zh_pruned_transducer_stateless2": _get_aidatatang_200zh_pretrained_mode,  # noqa
+    "luomingshuang/icefall_asr_alimeeting_pruned_transducer_stateless2": _get_alimeeting_pre_trained_model,  # noqa
 }
 
 english_models = {
     "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2": _get_gigaspeech_pre_trained_model,  # noqa
+    "WeijiZhuang/icefall-asr-librispeech-pruned-transducer-stateless8-2022-12-02": _get_librispeech_pre_trained_model,  # noqa
+    "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless8-2022-11-14": _get_librispeech_pre_trained_model,  # noqa
+    "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11": _get_librispeech_pre_trained_model,  # noqa
     "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13": _get_librispeech_pre_trained_model,  # noqa
 }
 
@@ -241,14 +290,21 @@ chinese_english_mixed_models = {
     "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5": _get_tal_csasr_pre_trained_model,  # noqa
 }
 
+tibetan_models = {
+    "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless7-2022-12-02": _get_tibetan_pre_trained_model,  # noqa
+    "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29": _get_tibetan_pre_trained_model,  # noqa
+}
+
 all_models = {
     **chinese_models,
     **english_models,
     **chinese_english_mixed_models,
+    **tibetan_models,
 }
 
 language_to_models = {
     "Chinese": list(chinese_models.keys()),
     "English": list(english_models.keys()),
     "Chinese+English": list(chinese_english_mixed_models.keys()),
+    "Tibetan": list(tibetan_models.keys()),
 }

requirements.txt

@@ -1,11 +1,9 @@
-https://download.pytorch.org/whl/cpu/torch-1.10.0%2Bcpu-cp38-cp38-linux_x86_64.whl
-https://k2-fsa.org/nightly/whl/k2-1.17.dev20220711+cpu.torch1.10.0-cp38-cp38-linux_x86_64.whl
-https://download.pytorch.org/whl/cpu/torchaudio-0.10.0%2Bcpu-cp38-cp38-linux_x86_64.whl
-
-
-https://huggingface.co/csukuangfj/wheels/resolve/main/kaldifeat-1.17-cp38-cp38-linux_x86_64.whl
-https://huggingface.co/csukuangfj/wheels/resolve/main/k2_sherpa-0.6-cp38-cp38-linux_x86_64.whl
+https://download.pytorch.org/whl/cpu/torch-1.13.0%2Bcpu-cp38-cp38-linux_x86_64.whl
+https://download.pytorch.org/whl/cpu/torchaudio-0.13.0%2Bcpu-cp38-cp38-linux_x86_64.whl
 
+https://huggingface.co/csukuangfj/wheels/resolve/main/k2-1.23.2.dev20221204%2Bcpu.torch1.13.0-cp38-cp38-linux_x86_64.whl
+https://huggingface.co/csukuangfj/wheels/resolve/main/kaldifeat-1.22-cp38-cp38-linux_x86_64.whl
+https://huggingface.co/csukuangfj/wheels/resolve/main/k2_sherpa-1.1-cp38-cp38-linux_x86_64.whl
 
 sentencepiece>=0.1.96
 numpy

test_wavs/tibetan/trans.txt

@@ -0,0 +1,3 @@
+a_0_cacm-A70_31116.wav ལོ བཅུ ཙམ མ འདང བའི དུས སྐབས ནང
+a_0_cacm-A70_31117.wav དྲག པོའི ངོ ལོག ཟིང འཁྲུག སྒྲིག འཛུགས དང ངན བཀོད བྱས ཡོད
+a_0_cacm-A70_31118.wav གནས བབ འདིའི རིགས གང མགྱོགས འགྱུར བ གཏོང དགོས