상황: 서버의 Jupyter 에서 Python 코드가 실행되지 않음
우측 상단의 Select Kernel에서 Python 3을 선택해도, 조금 뒤 No Kernel로 자동 전환
Shell들은 실행되지 않음

해결: 터미널 창에서 다음을 입력
conda install -n base ipykernel --update-deps --force-reinstall

실행문
python ./bin/main.py model=ds2 train=ds2_train train.dataset_path=/opt/ml/input/kspon_dataset/train train=ds2_train train.transcripts_path=/opt/ml/input/kospeech/dataset/kspon/transcripts.txt

에러 내용
OSError: /opt/conda/lib/python3.8/site-packages/torchaudio/_torchaudio.so: undefined symbol: _ZN3c104impl23ExcludeDispatchKeyGuardC1ENS_14DispatchKeySetE

get_model_binary.py 모델 실행 시 발생하는 오류

[오류]
TypeError: load() missing 1 required positional argument: 'Loader'

Title

• TypeError: recognize() takes 3 positional arguments but 4 were given

Description

recognize엔 3개의 arguments가 들어가야 하는데, 4개가 들어가고 있다.

opt.device를 사용하지 않으므로 빼 준다.

/opt/ml/input/kospeech/bin/inference.py

변경 전

if isinstance(model, ListenAttendSpell):
    model.encoder.device = opt.device
    model.decoder.device = opt.device

    y_hats = model.recognize(feature.unsqueeze(0), input_length, opt.device)
elif isinstance(model, DeepSpeech2):
    model.device = opt.device
    y_hats = model.recognize(feature.unsqueeze(0), input_length, opt.device)
elif isinstance(model, SpeechTransformer) or isinstance(model, Jasper) or isinstance(model, Conformer):
    y_hats = model.recognize(feature.unsqueeze(0), input_length, opt.device)

변경 후

if isinstance(model, ListenAttendSpell):
    model.encoder.device = opt.device
    model.decoder.device = opt.device

    y_hats = model.recognize(feature.unsqueeze(0), input_length)
elif isinstance(model, DeepSpeech2):
    model.device = opt.device
    y_hats = model.recognize(feature.unsqueeze(0), input_length)
elif isinstance(model, SpeechTransformer) or isinstance(model, Jasper) or isinstance(model, Conformer):
    y_hats = model.recognize(feature.unsqueeze(0), input_length)

Linked Issues

• resolved #

Title

• RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

Description

feature를 변경하여 해결

/opt/ml/input/kospeech/bin/inference.py

변경 전

elif isinstance(model, DeepSpeech2):
    model.device = opt.device
    y_hats = model.greedy_search(feature.unsqueeze(0), input_length, opt.device)

변경 후

elif isinstance(model, DeepSpeech2):
    model.device = opt.device
    feature = feature.unsqueeze(0).to(torch.device("cuda"))
    y_hats = model.recognize(feature, input_length)

Linked Issues

• resolved #

pip install numpy

pip install torch

pip install pandas

pip install matplotlib

conda install -c conda-forge librosa

pip install torchaudio==0.6.0

pip install tqdm

pip install warp_rnnt

pip install sentencepiece


pip install hydra-core --upgrade

pip install warp_rnnt는 CUDA error 발생

ModuleNotFoundError: No module named 'openspeech.data.text'
train.sh를 만들고, sh train.sh를 실행시켰으나 잘 되지 않았다.

해결
pip install -e .

❓ Questions & Help

@hydra.main(config_path=os.path.join("..", "openspeech", "configs"), config_name="train")
LexerNoViableAltException:
^
See https://hydra.cc/docs/next/advanced/override_grammar/basic for details

Details

train example 1을 실행하면 위와 같은 에러 발생

실행 내용 >
python ./bin/main.py model=ds2 train=ds2_train train.dataset_path=/opt/ml/input/kspon_dataset/train train.transcripts_path=/opt/ml/input/kospeech/dataset/kspon/transcripts.txt

오류 >
ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by /opt/conda/envs/kospeech/lib/python3.8/site-packages/scipy/linalg/_matfuncs_sqrtm_triu.cpython-38-x86_64-linux-gnu.so)

실행 코드 >
python ./openspeech_cli/hydra_train.py \ .....

오류 >
OSError: /opt/ml/input/env/lib/python3.8/site-packages/torchaudio/_torchaudio.so: undefined symbol: _ZNK2at6Tensor6deviceEv

Traceback (most recent call last):
File "./bin/inference.py", line 26, in
from kospeech.data.audio.core import load_audio
File "/opt/ml/input/kospeech/bin/kospeech/data/init.py", line 15, in
from kospeech.data.audio.parser import SpectrogramParser
File "/opt/ml/input/kospeech/bin/kospeech/data/audio/parser.py", line 18, in
from kospeech.data.audio.core import load_audio
File "/opt/ml/input/kospeech/bin/kospeech/data/audio/core.py", line 17, in
import librosa
File "/opt/conda/envs/kospeech/lib/python3.8/site-packages/librosa/init.py", line 209, in
from . import core
File "/opt/conda/envs/kospeech/lib/python3.8/site-packages/librosa/core/init.py", line 6, in
from .audio import * # pylint: disable=wildcard-import
File "/opt/conda/envs/kospeech/lib/python3.8/site-packages/librosa/core/audio.py", line 8, in
import soundfile as sf
File "/opt/conda/envs/kospeech/lib/python3.8/site-packages/soundfile.py", line 17, in
from _soundfile import ffi as _ffi
File "/opt/conda/envs/kospeech/lib/python3.8/site-packages/_soundfile.py", line 2, in
import _cffi_backend
ImportError: libffi.so.7: cannot open shared object file: No such file or directory

inference 실행에서 위와 같은 에러 발생

Title

• TypeError: init() got an unexpected keyword argument 'require'

Description

require는 없어서 생기는 오류

/opt/ml/input/kospeech/bin/inference.py

변경 전

parser.add_argument('--model_path', type=str, require=True)
parser.add_argument('--audio_path', type=str, require=True)
parser.add_argument('--device', type=str, require=False, default='cpu')

변경 후

parser.add_argument('--model_path', type=str, required=True)
parser.add_argument('--audio_path', type=str, required=True)
parser.add_argument('--device', type=str, required=False, default='cpu')

Linked Issues

• resolved #

Title

• ImportError: cannot import name 'BeamDecoderRNN' from 'kospeech.models.las.decoder' (/opt/ml/kospeech/kospeech/models/las/decoder.py)

Description

Before

from kospeech.models.las.decoder import DecoderRNN, BeamDecoderRNN

After

from kospeech.models.las.decoder import DecoderRNN

Linked Issues

• resolved #

터미널 입력
python ./bin/main.py model=ds2 train=ds2_train train.dataset_path=/opt/ml/input/kspon_dataset/train train.transcripts_path=/opt/ml/input/kospeech/dataset/kspon/transcripts.txt

오류명
File "/opt/conda/envs/stt/lib/python3.8/site-packages/soundfile.py", line 142, in
raise OSError('sndfile library not found')
OSError: sndfile library not found

Title

• AttributeError: 'DeepSpeech2' object has no attribute 'greedy_search’

Description

greedy_search도 없다고 한다.

/opt/ml/input/kospeech/bin/inference.py

변경 전

if isinstance(model, ListenAttendSpell):
    model.encoder.device = opt.device
    model.decoder.device = opt.device

    y_hats = model.greedy_search(feature.unsqueeze(0), input_length, opt.device)
elif isinstance(model, DeepSpeech2):
    model.device = opt.device
    y_hats = model.greedy_search(feature.unsqueeze(0), input_length, opt.device)
elif isinstance(model, SpeechTransformer) or isinstance(model, Jasper) or isinstance(model, Conformer):
    y_hats = model.greedy_search(feature.unsqueeze(0), input_length, opt.device)

변경 후

if isinstance(model, ListenAttendSpell):
    model.encoder.device = opt.device
    model.decoder.device = opt.device

    y_hats = model.recognize(feature.unsqueeze(0), input_length, opt.device)
elif isinstance(model, DeepSpeech2):
    model.device = opt.device
    y_hats = model.recognize(feature.unsqueeze(0), input_length, opt.device)
elif isinstance(model, SpeechTransformer) or isinstance(model, Jasper) or isinstance(model, Conformer):
    y_hats = model.recognize(feature.unsqueeze(0), input_length, opt.device)

Linked Issues

• resolved #