Tests crash on Ubuntu Jammy about espeak-phonemizer HOT 7 OPEN

The CLI doesn't segfault for me, but the tests certainly do. #3 fixes it so having that merged in a future release would be great!

from espeak-phonemizer.

Please upgrade to 1.3.1 and try again (thanks @msftcangoblowm)

from espeak-phonemizer.

With some print debugging, near line 100

        print("ptr", text_pointer)
        while text_pointer:
            clause_phonemes = ctypes.c_char_p(
                self.lib_espeak.espeak_TextToPhonemes(
                    ctypes.pointer(text_pointer), text_flags, phoneme_flags,
                )
            )
            print("ptr", text_pointer)
            print("clause", clause_phonemes)
            print("value", clause_phonemes.value)
            if clause_phonemes.value is not None:
                phoneme_lines.append(
                    clause_phonemes.value.decode()  # pylint: disable=no-member
                )

It prints:

$ echo "Hello" | python3 -m espeak_phonemizer -v en-us
ptr c_char_p(140501981572336)
ptr c_char_p(None)
clause c_char_p(2096739184)
Segmentation fault

The text is read (first the pointer exists, then it is None).
When value of clause_phonemes is get, it crashs.

from espeak-phonemizer.

Hi there, same problem here, three different systems.

1. On google colab, ok

2. On my system, Ubuntu 18.04 with python 3.7.6, Segmentation Fault

3. On Docker, using the image for training Piper, python 3.8.1.2 Segmentation Fault

From 2) I created a conda env with Python 3.9, it worked.

So it seems that the problem is Python <3.9 (Colab is shipped with the 3.9.16) .

EDIT: 3)from the Docker image "nvcr.io/nvidia/pytorch:22.03-py3" taken from the Piper repository, creating a conda env with python 3.9.16 still gives segmentation fault

from espeak-phonemizer.

Hi,

it appears, that a restype for an external function call must be set, otherwise python assumes, that the return type of the function is always a C int type.

so if in def _maybe_init(self):, after library loading something like this is done:

self.lib_espeakTTP = self.lib_espeak.espeak_TextToPhonemes
self.lib_espeakTTP.restype = ctypes.c_char_p

and then instead of using self.lib_espeak.espeak_TextToPhonemes directly in the phonemize function, use the self.lib_espeakTTP.
I'm not a python coder, so maybe there's another (cleaner) way of doing this, but this way appears to be working, if someone needs it right away...

from espeak-phonemizer.

Following methosm excellent advice.
The entire working method (minimal changes).

    def _phonemize_no_stream(self, text: str, \
        phoneme_separator: Optional[str]) -> List[str]:
        phoneme_flags = Phonemizer.espeakPHONEMES_IPA
        if phoneme_separator:
            phoneme_flags = phoneme_flags | (ord(phoneme_separator) << 8)

        text_bytes = text.encode("utf-8")
        text_pointer = ctypes.c_char_p(text_bytes)
        text_flags = Phonemizer.espeakCHARS_AUTO
        phoneme_lines = []
        fcn_ttp = self.lib_espeak.espeak_TextToPhonemes
        fcn_ttp.restype = ctypes.c_char_p
        while text_pointer:
            clause_phonemes = fcn_ttp( \
            ctypes.pointer(text_pointer), text_flags, \
            phoneme_flags)
            if isinstance(clause_phonemes, bytes):
                clause_phonemes_str = clause_phonemes.decode()  # pylint: disable=no-member
                phoneme_lines.append(clause_phonemes_str)

        return phoneme_lines

nefastosaturo suggests python 3.9.16 could be a cause
I have it working on python 3.9.16
This issue, mostly likely, has nothing to do with the python interpreter version

Edit
June 26th, 2023: this comment cleaned up; sorry about the typo. In espeak, Michael fixed that typo in a later commit

from espeak-phonemizer.

#3

from espeak-phonemizer.