Memory leak that consumes 5 MB per usage of phonemize function

[ollayf]

Describe the bug There is a memory leak where each pass of the phonemize function for me takes up at least 5 mb. For some reason I have tried many things but to no avail. This is how I use it in my python code

Phonemizer version

System Ubuntu 20.04 LTS Python 3.8

To reproduce

Expected behavior Everytime the function ends, the memory should be collected in the garbage and released back to the OS. But every time it runs it permanently takes up 5 MB. This 5MB is seen from when i use htop and when i use psutil.Process(os.getpid()).memory_info().rss / 1024 ** 2 in the python code

Additional context

[mmmaat]

Hi,

See here you may try passing a backend to your phonemize_word function. When creating a new espeak phonemization instance, the code actually copies the espeak shared library somewhere in a temp directory (that's the 5mb). Normally the directory is deleted at exit or when garbage collected (see here this is a bit complex to deal with Linux/Mac/Windows).

[magicse]

Same problem leak memory on every import phonemizer

phonemes = phonemizer.phonemize({orig_text_wo_stress}, language="en")

[lokmantsui]

don't instantiate one phonemizer backend per call https://github.com/bootphon/phonemizer/blob/d9f9ed266aa5cc2dd9e5eaea2c9571ab5024893c/phonemizer/phonemize.py#L206