Update app.py

app.py

@@ -20,39 +20,42 @@ from datasets import set_caching_enabled
 set_caching_enabled(False)
 disable_caching()
 
-from transformers import pipeline, WhisperModel, WhisperTokenizer, AutoConfig
-from datasets import load_dataset
+p = pipeline("automatic-speech-recognition")
 
+#config = AutoConfig.from_pretrained('whisper-small')
 
-metric = evaluate.load("wer")
-
-# Load the Whisper model and tokenizer
 huggingface_token = os.environ["huggingface_token"]
-whisper_miso =AutoModelForCTC.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
-miso_tokenizer = AutoModelForCTC.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
-
-# Initialize the automatic-speech-recognition pipeline with the Whisper model and tokenizer
-asr_pipeline = pipeline(
-    "automatic-speech-recognition",
-    model=whisper_miso,
-    tokenizer=miso_tokenizer
-)
 
-# Load the dataset
-dataset = load_dataset("mskov/miso_test", split="test").cast_column("audio", Audio())
+whisper_miso=WhisperModel.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
+miso_tokenizer = WhisperTokenizer.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
+#miso_tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-small", use_auth_token=huggingface_token)
 
-# Compute the evaluation results
-results = asr_pipeline(dataset)
-metric = WERMetric()
-wer = metric.compute(predictions=results, references=dataset["audio"])
-print(wer)
+task_evaluator = p.task_evaluator
+task_evaluator.feature_extractor = whisper_miso.get_feature_extractor()
 
+task_evaluator = evaluator("automatic-speech-recognition")
+#url = {"test" : "https://huggingface.co/datasets/mskov/miso_test/blob/main/test_set.parquet"}
+#data = load_dataset("audiofolder", data_dir="mskov/miso_test")
+# data = load_dataset("audiofolder", data_files=["datasets/mskov/miso_test/test_set/and.wav","mskov/miso_test/test_set/chew1.wav","mskov/miso_test/test_set/chew3.wav", "mskov/miso_test/test_set/chew3.wav","mskov/miso_test/test_set/chew4.wav","mskov/miso_test/test_set/cough1.wav","mskov/miso_test/test_set/cough2.wav","mskov/miso_test/test_set/cough3.wav","mskov/miso_test/test_set/hi.wav","mskov/miso_test/test_set/knock_knock.wav","mskov/miso_test/test_set/mouth_sounds1.wav","mskov/miso_test/test_set/mouth_sounds2.wav","mskov/miso_test/test_set/no.wav","mskov/miso_test/test_set/not_bad.wav","mskov/miso_test/test_set/oh_i_wish.wav","mskov/miso_test/test_set/pop1.wav","mskov/miso_test/test_set/really.wav","mskov/miso_test/test_set/sigh1.wav","mskov/miso_test/test_set/sigh2.wav","mskov/miso_test/test_set/slurp1.wav","mskov/miso_test/test_set/slurp2.wav","mskov/miso_test/test_set/sneeze1.wav","mskov/miso_test/test_set/sneeze2.wav","mskov/miso_test/test_set/so_i_did_it_again.wav"])
+dataset = load_dataset("mskov/miso_test", split="test").cast_column("audio", Audio())
+results = task_evaluator.compute(
+    model_or_pipeline=whisper_miso,
+    #model_or_pipeline="mskov/whisper-small.en",
+    data=dataset,
+    tokenizer=miso_tokenizer,
+    input_column="audio",
+    label_column="audio",
+    # device=None,
+    strategy="simple",
+    metric="wer",
+)
+print(results)
 
 
 def transcribe(audio, state=""):
     text = p(audio)["text"]
     state += text + " "
-    return state, state
+    returnstate, state
 
 gr.Interface(
     fn=transcribe, 
@@ -64,4 +67,4 @@ gr.Interface(
         "textbox",
         "state"
     ],
-    live=True).launch()
+    live=True).launch()