whisper-large_v2_test / handler.py
slplab's picture
Update handler.py
0f9c5ae
raw
history blame
740 Bytes
from typing import Dict, Any, List
from transformers import pipeline
import torch
from transformers.pipelines.audio_utils import ffmpeg_read
#ffmpeg
class EndpointHandler:
def __init__(self, path=""):
self.pipe = pipeline(task='automatic-speech-recognition', model=path)
def __call__(self, data: Any) -> List[Dict[str, str]]:
inputs = data.pop("inputs", data)
audio_nparray = ffmpeg_read(inputs, 16000)
audio_tensor= torch.from_numpy(audio_nparray)
transcribe = self.pipe
transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ko", task="transcribe")
result = transcribe(audio_nparray)
return result