Update app.py
Browse files
app.py
CHANGED
@@ -18,36 +18,31 @@ from datasets import set_caching_enabled
|
|
18 |
set_caching_enabled(False)
|
19 |
disable_caching()
|
20 |
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
|
|
|
25 |
huggingface_token = os.environ["huggingface_token"]
|
26 |
-
|
27 |
-
whisper_miso=WhisperModel.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
|
28 |
miso_tokenizer = WhisperTokenizer.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
|
29 |
-
#miso_tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-small", use_auth_token=huggingface_token)
|
30 |
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
|
|
33 |
|
34 |
-
|
35 |
-
#url = {"test" : "https://huggingface.co/datasets/mskov/miso_test/blob/main/test_set.parquet"}
|
36 |
-
#data = load_dataset("audiofolder", data_dir="mskov/miso_test")
|
37 |
-
# 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"])
|
38 |
dataset = load_dataset("mskov/miso_test", split="test").cast_column("audio", Audio())
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
# device=None,
|
47 |
-
strategy="simple",
|
48 |
-
metric="wer",
|
49 |
-
)
|
50 |
-
print(results)
|
51 |
|
52 |
|
53 |
def transcribe(audio, state=""):
|
|
|
18 |
set_caching_enabled(False)
|
19 |
disable_caching()
|
20 |
|
21 |
+
from transformers import pipeline, WhisperModel, WhisperTokenizer, AutoConfig
|
22 |
+
from datasets import load_dataset
|
23 |
+
from transformers import WERMetric
|
24 |
|
25 |
+
# Load the Whisper model and tokenizer
|
26 |
huggingface_token = os.environ["huggingface_token"]
|
27 |
+
whisper_miso = WhisperModel.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
|
|
|
28 |
miso_tokenizer = WhisperTokenizer.from_pretrained("mskov/whisper_miso", use_auth_token=huggingface_token)
|
|
|
29 |
|
30 |
+
# Initialize the automatic-speech-recognition pipeline with the Whisper model and tokenizer
|
31 |
+
asr_pipeline = pipeline(
|
32 |
+
"automatic-speech-recognition",
|
33 |
+
model=whisper_miso,
|
34 |
+
tokenizer=miso_tokenizer
|
35 |
+
)
|
36 |
|
37 |
+
# Load the dataset
|
|
|
|
|
|
|
38 |
dataset = load_dataset("mskov/miso_test", split="test").cast_column("audio", Audio())
|
39 |
+
|
40 |
+
# Compute the evaluation results
|
41 |
+
results = asr_pipeline(dataset)
|
42 |
+
metric = WERMetric()
|
43 |
+
wer = metric.compute(predictions=results, references=dataset["audio"])
|
44 |
+
print(wer)
|
45 |
+
|
|
|
|
|
|
|
|
|
|
|
46 |
|
47 |
|
48 |
def transcribe(audio, state=""):
|