File size: 3,537 Bytes
e68ccbb ef9a4d0 e68ccbb ef9a4d0 e68ccbb ef9a4d0 e68ccbb ef9a4d0 a3bb507 e68ccbb ef9a4d0 709c5f5 e68ccbb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 |
from typing import Dict, Any, List
from transformers import WhisperForConditionalGeneration, AutoProcessor, WhisperTokenizer, WhisperProcessor, pipeline, WhisperFeatureExtractor
import torch
#import io
#device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
class EndpointHandler:
def __init__(self, path=""):
#tokenizer = WhisperTokenizer.from_pretrained('openai/whisper-large', language="korean", task='transcribe')
#model = WhisperForConditionalGeneration.from_pretrained(path)
#self.tokenizer = WhisperTokenizer.from_pretrained(path)
#self.processor = WhisperProcessor.from_pretrained(path, language="korean", task='transcribe')
#processor = AutoProcessor.from_pretrained(path)
#self.pipe = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.feature_extractor, feature_extractor=processor.feature_extractor)
#feature_extractor = WhisperFeatureExtractor.from_pretrained('openai/whisper-large')
self.pipe = pipeline(task='automatic-speech-recognition', model=path)
# Move model to device
# self.model.to(device)
def __call__(self, data: Any) -> List[Dict[str, str]]:
print('==========NEW PROCESS=========')
transcribe = pipeline(task="automatic-speech-recognition", model="vasista22/whisper-kannada-tiny")
transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ko", task="transcribe")
result = transcribe(data['inputs'])
#print(f"{data}")
#inputs = data.pop("inputs", data)
#print(f'1. inputs: {inputs}')
#inputs, _ = sf.read(io.BytesIO(data['inputs']))
#inputs, _ = sf.read(data['inputs'])
#print(f'2. inputs: {inputs}')
# input_features = self.feature_extractor(inputs, sampling_rate=16000).input_features[0]
# #print(f'3. input_features: {input_features}')
# input_features_tensor = torch.tensor(input_features).unsqueeze(0)
# input_ids = self.model.generate(input_features_tensor)
# #(f'4. input_ids: {input_ids}')
# transcription = self.tokenizer.batch_decode(input_ids, skip_special_tokens=True)[0]
# #inputs, _ = torchaudio.load(inputs, normalize=True)
# #input_features = self.processor.feature_extractor(inputs, sampling_rate=16000).input_features[0]
#input_ids = self.processor.tokenizer(input_features, return_tensors="pt").input_ids
#generated_ids = self.model.generate(input_ids)
# #transcription = self.pipe(inputs, generate_kwargs = {"task":"transcribe", "language":"<|ko|>"})
# #transcription = self.pipe(inputs)
# #print(input)
# inputs = self.processor(inputs, retun_tensors="pt")
# #input_features = {key: value.to(device) for key, value in input_features.items()}
# input_features = inputs.input_features
# generated_ids = self.model.generate(input_features)
# #generated_ids = self.model.generate(inputs=input_features)
# #self.model.generate = partial(self.model.generate, language="korean", task="transcribe")
# #generated_ids = self.model.generate(inputs = input_features)
#transcription = self.processor.tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
#transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
return result |