Delete handler.py
Browse files- handler.py +0 -33
handler.py
DELETED
|
@@ -1,33 +0,0 @@
|
|
| 1 |
-
from typing import Dict
|
| 2 |
-
from transformers.pipelines.audio_utils import ffmpeg_read
|
| 3 |
-
import whisper
|
| 4 |
-
import torch
|
| 5 |
-
|
| 6 |
-
SAMPLE_RATE = 16000
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
class EndpointHandler():
|
| 11 |
-
def __init__(self, path=""):
|
| 12 |
-
# load the model
|
| 13 |
-
self.model = whisper.load_model("medium")
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
def __call__(self, data: Dict[str, bytes]) -> Dict[str, str]:
|
| 17 |
-
"""
|
| 18 |
-
Args:
|
| 19 |
-
data (:obj:):
|
| 20 |
-
includes the deserialized audio file as bytes
|
| 21 |
-
Return:
|
| 22 |
-
A :obj:`dict`:. base64 encoded image
|
| 23 |
-
"""
|
| 24 |
-
# process input
|
| 25 |
-
inputs = data.pop("inputs", data)
|
| 26 |
-
audio_nparray = ffmpeg_read(inputs, SAMPLE_RATE)
|
| 27 |
-
audio_tensor= torch.from_numpy(audio_nparray)
|
| 28 |
-
|
| 29 |
-
# run inference pipeline
|
| 30 |
-
result = self.model.transcribe(audio_nparray)
|
| 31 |
-
|
| 32 |
-
# postprocess the prediction
|
| 33 |
-
return {"text": result["text"]}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|