Views
No views yet
1>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
2>>> import librosa
3
4>>> # load model and processor
5>>> processor = WhisperProcessor.from_pretrained("akuzdeuov/whisper-base.kk")
6>>> model = WhisperForConditionalGeneration.from_pretrained("akuzdeuov/whisper-base.kk")
7
8>>> # load your audio
9>>> audio, sampling_rate = librosa.load("path_to_audio", sr=16000)
10>>> input_features = processor(audio, sampling_rate=sampling_rate, return_tensors="pt").input_features
11
12>>> # generate token ids
13>>> predicted_ids = model.generate(input_features)
14>>> # decode token ids to text
15>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=False)
16
17>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)skip_special_tokens=True.pipeline
method. Chunking is enabled by setting chunk_length_s=30 when instantiating the pipeline. With chunking enabled, the pipeline
can be run with batched inference.1>>> import torch
2>>> from transformers import pipeline
3
4>>> device = "cuda:0" if torch.cuda.is_available() else "cpu"
5
6>>> pipe = pipeline(
7>>> "automatic-speech-recognition",
8>>> model="akuzdeuov/whisper-base.kk",
9>>> chunk_length_s=30,
10>>> device=device,
11>>> )
12
13>>> prediction = pipe("path_to_audio", batch_size=8)["text"]