def _asr(model_name):
d = ModelDownloader("downloads")
speech2text = Speech2Text(**d.download_and_unpack(model_name, quiet=True))
speech = np.zeros((10000, ), dtype=np.float32)
nbests = speech2text(speech)
text, *_ = nbests[0]
assert isinstance(text, str)
def __init__(self, model_name, trans_df):
from espnet2.bin.asr_inference import Speech2Text
from espnet_model_zoo.downloader import ModelDownloader
import jiwer
self.model_name = model_name
d = ModelDownloader()
self.asr_model = Speech2Text(**d.download_and_unpack(model_name))
self.input_txt_list = []
self.clean_txt_list = []
self.output_txt_list = []
self.transcriptions = []
self.true_txt_list = []
self.sample_rate = int(
d.data_frame[d.data_frame["name"] == model_name]["fs"])
self.trans_df = trans_df
self.trans_dic = self._df_to_dict(trans_df)
self.mix_counter = Counter()
self.clean_counter = Counter()
self.est_counter = Counter()
self.transformation = jiwer.Compose([
jiwer.ToLowerCase(),
jiwer.RemovePunctuation(),
jiwer.RemoveMultipleSpaces(),
jiwer.Strip(),
jiwer.SentencesToListOfWords(),
jiwer.RemoveEmptyStrings(),
])
def get_speech2text():
d = ModelDownloader()
speech2text = Speech2Text(
# Specify task and corpus
# **d.download_and_unpack(task="asr", corpus="librispeech")
**d.download_and_unpack(
"Shinji Watanabe/laborotv_asr_train_asr_conformer2_latest33_raw_char_sp_valid.acc.ave"
),
device='cuda')
return speech2text
def test_Speech2Text(asr_config_file, lm_config_file):
speech2text = Speech2Text(asr_train_config=asr_config_file,
lm_train_config=lm_config_file,
beam_size=1)
speech = np.random.randn(100000)
results = speech2text(speech)
for text, token, token_int, hyp in results:
assert isinstance(text, str)
assert isinstance(token[0], str)
assert isinstance(token_int[0], int)
assert isinstance(hyp, Hypothesis)
def test_EnhS2T_Speech2Text(enh_asr_config_file, lm_config_file):
speech2text = Speech2Text(
asr_train_config=enh_asr_config_file,
lm_train_config=lm_config_file,
beam_size=1,
enh_s2t_task=True,
)
speech = np.random.randn(48000)
results = speech2text(speech)
for text, token, token_int, hyp in results:
assert isinstance(text, str)
assert isinstance(token[0], str)
assert isinstance(token_int[0], int)
assert isinstance(hyp, Hypothesis)
def __init__(self, model_name, trans_df):
from espnet2.bin.asr_inference import Speech2Text
from espnet_model_zoo.downloader import ModelDownloader
self.model_name = model_name
d = ModelDownloader()
self.asr_model = Speech2Text(**d.download_and_unpack(model_name))
self.input_txt_list = []
self.clean_txt_list = []
self.output_txt_list = []
self.sample_rate = int(
d.data_frame[d.data_frame["name"] == model_name]["fs"])
self.trans_df = trans_df
self.trans_dic = self._df_to_dict(trans_df)
self.mix_counter = Counter()
self.clean_counter = Counter()
self.est_counter = Counter()
import Levenshtein
import json
import warnings
from pathlib import Path
import hydra.utils as utils
import matplotlib.pyplot as plt
import pickle
import seaborn as sns
from utils import normalize
sns.set()
warnings.simplefilter('ignore')
d = ModelDownloader()
speech2text_en = Speech2Text(
**d.download_and_unpack(task="asr", corpus="librispeech")
)
def log_spec_dB_dist(x, y):
log_spec_dB_const = 10.0 / math.log(10.0) * math.sqrt(2.0)
diff = x - y
return log_spec_dB_const * math.sqrt(np.inner(diff, diff))
def average_mcd(mc_ref, mc_cv):
_, path = fastdtw(mc_cv, mc_ref, dist=euclidean)
twf = np.array(path).T
cvt_mcc_dtw = mc_cv[twf[0]]
trg_mcc_dtw = mc_ref[twf[1]]
# MCD
diff2sum = np.sum((cvt_mcc_dtw - trg_mcc_dtw)**2, 1)
mcd_value = np.mean(10.0 / np.log(10.0) * np.sqrt(2 * diff2sum), 0)
def __init__(self, model_id: str):
self.model = Speech2Text.from_pretrained(model_id, device="cpu", beam_size=1)
self.sampling_rate = 16000