Can't use multiple EspeakBackend objects with njobs=1

[Ishaan Kumar]

Describe the bug
It seems that instantiation of multiple EspeakBackend objects is not correctly handled.
All the objects start operating with the language used to instantiate the last object.
Please refer to the example below.

Phonemizer version
3.0.1

System
macOS 11.6.4
python 3.8.9
[Clang 13.0.0 (clang-1300.0.29.30)] on darwin

To reproduce

from phonemizer.backend import EspeakBackend

en_backend = EspeakBackend(
    "en-us",
    preserve_punctuation=True,
    with_stress=True,
    language_switch="remove-flags",
    words_mismatch="ignore",
)
en_sentence = ["I love to eat pizza everyday"]
print(en_backend.phonemize(en_sentence, njobs=1, strip=True)) # ['aɪ lˈʌv tʊ ˈiːt pˈiːtsə ˈɛvɹɪdˌeɪ']

de_backend = EspeakBackend(
    "de",
    preserve_punctuation=True,
    with_stress=True,
    language_switch="remove-flags",
    words_mismatch="ignore",
)
de_sentence = ["ich esse jeden tag gerne pizza."]
print(de_backend.phonemize(de_sentence, njobs=1, strip=True)) # ['ɪç ˈɛsə jˈeːdən tˈɑːk ɡˈɛɾnə pˈɪtsɑː.']

incorrect_en = en_backend.phonemize(en_sentence, njobs=1, strip=True)
en_with_de = de_backend.phonemize(en_sentence, njobs=1, strip=True)

assert en_with_de == incorrect_en
print(incorrect_en, en_with_de) 
# ['ˈiː lˈoːvə tˈoː eːˈɑːt pˈɪtsɑː ˈeːveːrˌyːdɛɪ'] 
# ['ˈiː lˈoːvə tˈoː eːˈɑːt pˈɪtsɑː ˈeːveːrˌyːdɛɪ']

Expected behavior
Notice that incorrect_en is equal to en_with_de and not equal to en_sentence.

Additional context
This problem happens only with njobs=1 and doesn't appear with njobs>1