58 lines
1.6 KiB
Python
58 lines
1.6 KiB
Python
|
import json
|
||
|
|
import os
|
||
|
|
|
||
|
|
from appPublic.dictObject import DictObject
|
||
|
|
from appPublic.zmq_reqrep import ZmqReplier
|
||
|
|
from appPublic.jsonConfig import getConfig
|
||
|
|
|
||
|
|
from time import time
|
||
|
|
from faster_whisper import WhisperModel
|
||
|
|
|
||
|
|
class ZmqASR:
|
||
|
|
def __init__(self):
|
||
|
|
config = getConfig()
|
||
|
|
self.model_dir = config.model_dir or "large-v3"
|
||
|
|
self.device = config.device or "cpu"
|
||
|
|
self.zmq_url = config.zmq_url
|
||
|
|
self.compute_type=config.compute_type or "float16"
|
||
|
|
self.model = WhisperModel(self.model_dir,
|
||
|
|
device=self.device,
|
||
|
|
compute_type=self.compute_type)
|
||
|
|
self.replier = ZmqReplier(self.zmq_url, self.generate)
|
||
|
|
|
||
|
|
def run(self):
|
||
|
|
print(f'running {self.zmq_url}')
|
||
|
|
self.replier._run()
|
||
|
|
print('ended ...')
|
||
|
|
|
||
|
|
def generate(self, d):
|
||
|
|
msg= d.decode('utf-8')
|
||
|
|
data = DictObject(**json.loads(msg))
|
||
|
|
t1 = time()
|
||
|
|
segs, info = self.model.transcribe(data.audio_file)
|
||
|
|
t2 = time()
|
||
|
|
d = DictObject(**{
|
||
|
|
'segments':[],
|
||
|
|
'language':info.language
|
||
|
|
})
|
||
|
|
d.content = ''
|
||
|
|
for s in segs:
|
||
|
|
seg = {
|
||
|
|
"start":s.start,
|
||
|
|
"end":s.end,
|
||
|
|
"text":s.text
|
||
|
|
}
|
||
|
|
d.segments.append(seg)
|
||
|
|
d.content += s.text
|
||
|
|
d.time_cost = t2 - t1
|
||
|
|
print(f'{d}')
|
||
|
|
return json.dumps(d)
|
||
|
|
|
||
|
|
if __name__ == '__main__':
|
||
|
|
workdir = os.getcwd()
|
||
|
|
config = getConfig(workdir)
|
||
|
|
print(f'{config=}')
|
||
|
|
asr = ZmqASR()
|
||
|
|
print('here')
|
||
|
|
asr.run()
|