-
Notifications
You must be signed in to change notification settings - Fork 0
/
resources.py
32 lines (23 loc) · 1.57 KB
/
resources.py
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
from config import *
import speech_recognition as sr
from speechbrain.pretrained import EncoderClassifier
from transformers.utils import logging as logging_transformers
logging_transformers.get_logger().setLevel(logging_transformers.CRITICAL)
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, BlenderbotForConditionalGeneration, BlenderbotTokenizer, AutoTokenizer,AutoModelForSeq2SeqLM
import pickle
from fairseq.checkpoint_utils import load_model_ensemble_and_task
from transformers import pipeline
id2label = pickle.load(open(id2label_path, 'rb'))
eng_processor = Wav2Vec2Processor.from_pretrained(eng_model_path)
eng_model = Wav2Vec2ForCTC.from_pretrained(eng_model_path)
hi_processor = Wav2Vec2Processor.from_pretrained(hi_model_path)
hi_model = Wav2Vec2ForCTC.from_pretrained(hi_model_path)
speech_models, cfg, task = load_model_ensemble_and_task(tts_model_path, arg_overrides={"config_yaml": "./config.yaml", "data": tts_path, "vocoder": "hifigan", "fp16": False})
speech_generator = task.build_generator(speech_models, cfg)
mt_tokenizer = AutoTokenizer.from_pretrained(mt_model_path)
mt_model = AutoModelForSeq2SeqLM.from_pretrained(mt_model_path)
conversational_tokenizer = BlenderbotTokenizer.from_pretrained(coversational_model_path)
conversational_model = BlenderbotForConditionalGeneration.from_pretrained(coversational_model_path)
speech_classifier = EncoderClassifier.from_hparams(source=speech_classifier_path)
restore_punct_model = pipeline("ner", model=restore_punct_model_path, tokenizer=restore_punct_model_path, aggregation_strategy="average")
recorder = sr.Recognizer()