ModuleNotFoundError: No module named 'ctc_decoders'

[KyuDan1]

File "/home/mid01/anaconda3/envs/py3.8/lib/python3.8/site-packages/nemo/collections/asr/modules/beam_search_decoder.py", line 64, in init
from ctc_decoders import Scorer, ctc_beam_search_decoder_batch
ModuleNotFoundError: No module named 'ctc_decoders'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "decoder_testing.py", line 38, in
beam_search = nemo_asr.modules.BeamSearchDecoderWithLM(
File "/home/mid01/anaconda3/envs/py3.8/lib/python3.8/site-packages/nemo/collections/asr/modules/beam_search_decoder.py", line 66, in init
raise ModuleNotFoundError(
ModuleNotFoundError: BeamSearchDecoderWithLM requires the installation of ctc_decoders from scripts/asr_language_modeling/ngram_lm/install_beamsearch_decoders.sh

My python version is 3.8 as #6690 said.

And I also tried

conda create --name nemo python==3.8.10
conda activate nemo

# install NeMo
sudo apt-get update && apt-get install -y libsndfile1 ffmpeg
pip install Cython
git clone https://github.com/NVIDIA/NeMo
cd NeMo
./reinstall.sh

# install ctc-decoders
./scripts/asr_language_modeling/ngram_lm/install_beamsearch_decoders.sh

but
./scripts/asr_language_modeling/ngram_lm/install_beamsearch_decoders.sh: line 26: cd: /workspace/nemo: No such file or directory
error occured then I get GPG error, No module named 'swig_decoders' ..

still doesn't work.

I just want to get BeamSearchDecoder's result.
Please help me.

[karpnv]

you need to provide a path to NeMo as a first argument of install_beamsearch_decoders.sh script:
./scripts/asr_language_modeling/ngram_lm/install_beamsearch_decoders.sh /path/to/your/nemo

[titu1994]

We need to catch these missing inputs and provide proper error message

[KyuDan1]

I typed ./scripts/asr_language_modeling/ngram_lm/install_beamsearch_decoders.sh /path/to/your/nemo
and successfully installed decoders.

But I still have some issues..

This is my code.

import nemo
import nemo.collections.asr as nemo_asr
import numpy as np
import os

am_model_quartznet = nemo_asr.models.ASRModel.from_pretrained(
    model_name="stt_en_quartznet15x5", strict=False
)

print("------------------am_model_loaded-------------------")


# softmax implementation in NumPy
def softmax(logits):
    e = np.exp(logits - np.max(logits))
    return e / e.sum(axis=-1).reshape([logits.shape[0], 1])


print("------------------def-------------------")

# Define number of CPUs to use. Set to the max when processing large batches of log probabilities
num_cpus = max(os.cpu_count(), 1)
print("------------------num_cpus-------------------")
# Set the beam size
beam_size = 16

# Get the vocabulary size
vocab = list(am_model_quartznet.decoder.vocabulary)
print("------------------vocab-------------------")

# Beam search
beam_search = nemo_asr.modules.BeamSearchDecoderWithLM(
    beam_width=beam_size,
    lm_path=None,
    alpha=None,
    beta=None,
    vocab=vocab,
    num_cpus=num_cpus,
    input_tensor=False,
)
print("------------------Beam search-------------------")
AUDIO_FILENAME = (
    "/home/introai1/Project/resampled_splitted_audio/audio_1-Orientation/output1.wav"
)
files = [AUDIO_FILENAME]
logits = am_model_quartznet.transcribe(files, logprobs=True)[0]

print("------------------logit-------------------")

probs = softmax(logits)
print("------------------probs-------------------")

best_sequences = beam_search.forward(
    log_probs=np.expand_dims(probs, axis=0), log_probs_length=None
)
print("Number of best sequences :", len(best_sequences[0]))
print("Best sequences :")
print(best_sequences[0])

And output was

[NeMo I 2023-08-26 00:53:47 features:289] PADDING: 16
[NeMo I 2023-08-26 00:53:48 save_restore_connector:249] Model EncDecCTCModel was successfully restored from /home/introai1/.cache/torch/NeMo/NeMo_1.21.0rc0/stt_en_quartznet15x5/16661021d16e679bdfd97a2a03944c49/stt_en_quartznet15x5.nemo.
------------------am_model_loaded-------------------
------------------def-------------------
------------------num_cpus-------------------
------------------vocab-------------------

It doesn't mean just error but I waited about 2 hours and nothing happened..
I guess something's wrong.

[KyuDan1]

I checked this error solved in python 3.10.
above was python 3.8.
Thank you!!