Update handler.py
Browse files- handler.py +6 -6
handler.py
CHANGED
@@ -15,17 +15,17 @@ class EndpointHandler:
|
|
15 |
self.tokenizer = WhisperTokenizer.from_pretrained('openai/whisper-large', language="korean", task='transcribe')
|
16 |
self.model = WhisperForConditionalGeneration.from_pretrained(path)
|
17 |
#self.tokenizer = WhisperTokenizer.from_pretrained(path)
|
18 |
-
self.processor = WhisperProcessor.from_pretrained(path, language="korean", task='transcribe')
|
19 |
#self.processor = AutoProcessor.from_pretrained(path)
|
20 |
#self.pipe = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.feature_extractor, feature_extractor=processor.feature_extractor)
|
21 |
-
self.feature_extractor = WhisperFeatureExtractor.from_pretrained(
|
22 |
|
23 |
|
24 |
# Move model to device
|
25 |
# self.model.to(device)
|
26 |
|
27 |
def __call__(self, data: Any) -> List[Dict[str, str]]:
|
28 |
-
print('
|
29 |
#print(f"{data}")
|
30 |
#inputs = data.pop("inputs", data)
|
31 |
#print(f'1. inputs: {inputs}')
|
@@ -33,13 +33,13 @@ class EndpointHandler:
|
|
33 |
|
34 |
inputs, _ = sf.read(io.BytesIO(data['inputs']))
|
35 |
#inputs, _ = sf.read(data['inputs'])
|
36 |
-
print(f'2. inputs: {inputs}')
|
37 |
|
38 |
input_features = self.feature_extractor(inputs, sampling_rate=16000).input_features[0]
|
39 |
-
print(f'3. input_features: {input_features}')
|
40 |
input_features_tensor = torch.tensor(input_features).unsqueeze(0)
|
41 |
input_ids = self.model.generate(input_features_tensor)
|
42 |
-
|
43 |
|
44 |
transcription = self.tokenizer.batch_decode(input_ids, skip_special_tokens=True)[0]
|
45 |
|
|
|
15 |
self.tokenizer = WhisperTokenizer.from_pretrained('openai/whisper-large', language="korean", task='transcribe')
|
16 |
self.model = WhisperForConditionalGeneration.from_pretrained(path)
|
17 |
#self.tokenizer = WhisperTokenizer.from_pretrained(path)
|
18 |
+
#self.processor = WhisperProcessor.from_pretrained(path, language="korean", task='transcribe')
|
19 |
#self.processor = AutoProcessor.from_pretrained(path)
|
20 |
#self.pipe = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.feature_extractor, feature_extractor=processor.feature_extractor)
|
21 |
+
self.feature_extractor = WhisperFeatureExtractor.from_pretrained('openai/whisper-large')
|
22 |
|
23 |
|
24 |
# Move model to device
|
25 |
# self.model.to(device)
|
26 |
|
27 |
def __call__(self, data: Any) -> List[Dict[str, str]]:
|
28 |
+
print('==========NEW PROCESS=========')
|
29 |
#print(f"{data}")
|
30 |
#inputs = data.pop("inputs", data)
|
31 |
#print(f'1. inputs: {inputs}')
|
|
|
33 |
|
34 |
inputs, _ = sf.read(io.BytesIO(data['inputs']))
|
35 |
#inputs, _ = sf.read(data['inputs'])
|
36 |
+
#print(f'2. inputs: {inputs}')
|
37 |
|
38 |
input_features = self.feature_extractor(inputs, sampling_rate=16000).input_features[0]
|
39 |
+
#print(f'3. input_features: {input_features}')
|
40 |
input_features_tensor = torch.tensor(input_features).unsqueeze(0)
|
41 |
input_ids = self.model.generate(input_features_tensor)
|
42 |
+
#(f'4. input_ids: {input_ids}')
|
43 |
|
44 |
transcription = self.tokenizer.batch_decode(input_ids, skip_special_tokens=True)[0]
|
45 |
|