Writing features as 'ark,scp' by pipeline with 'copy-feats'

[faber6911]

Hi, I can't correctly execute the example you provided to write the .ark and .scp files at the same time. If instead I create the ark file and use copy-feats to create its copy and the attached .scp file, I don't encounter any problems.

[jensen199105]

i wanna ask the the same thing, here is my code snippet and errors:

`import numpy as np from kaldiio import ReadHelper import kaldi_io

ark_scp_output = 'ark:| copy-feats ark:- ark,scp:ark:/home/jensen/Document/feats.ark,scp:/home/jensen/Document/feats.scp'

with kaldi_io.open_or_fd(ark_scp_output, 'w') as f: dic = {} for i in range(10): arr = np.random.randn(200, 10) dic[str(i)] = arr for k,v in dic.items(): kaldi_io.write_mat(f, v, k)` copy-feats ark:- ark,scp:ark:/home/jensen/Document/feats.ark,scp:/home/jensen/Document/feats.scp WARNING (copy-feats[5.5.839~1-0c6a]:Open():util/kaldi-table-inl.h:1311) When writing to both archive and script, the script file will generally not be interpreted correctly unless the archive is an actual file: wspecifier = ark,scp:ark:/home/jensen/Document/feats.ark,scp:/home/jensen/Document/feats.scp WARNING (copy-feats[5.5.839~1-0c6a]:Open():kaldi-io.cc:729) Invalid output filename format ark:/home/jensen/Document/feats.ark ERROR (copy-feats[5.5.839~1-0c6a]:TableWriter():util/kaldi-table-inl.h:1469) Failed to open table for writing with wspecifier: ark,scp:ark:/home/jensen/Document/feats.ark,scp:/home/jensen/Document/feats.scp: errno (in case it's relevant) is: Success

[ Stack-Trace: ] copy-feats(kaldi::MessageLogger::LogMessage() const+0x77b) [0x561ab18e1275] copy-feats(kaldi::MessageLogger::LogAndThrow::operator=(kaldi::MessageLogger const&)+0x25) [0x561ab1852001] copy-feats(kaldi::TableWriter<kaldi::KaldiObjectHolder<kaldi::MatrixBase > >::TableWriter(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&)+0xee) [0x561ab1861ba6] copy-feats(main+0x4c9) [0x561ab184ff92] /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf3) [0x7f82d5ae10b3] copy-feats(_start+0x2e) [0x561ab184fa0e]

kaldi::KaldiFatalErrorException in thread Thread-1: Traceback (most recent call last): File "/usr/lib/python3.8/threading.py", line 932, in _bootstrap_inner self.run() File "/usr/lib/python3.8/threading.py", line 870, in run self._target(*self._args, **self._kwargs) File "/home/jensen/.local/lib/python3.8/site-packages/kaldi_io/kaldi_io.py", line 97, in cleanup raise SubprocessFailed('cmd %s returned %d !' % (cmd,ret)) kaldi_io.kaldi_io.SubprocessFailed: cmd copy-feats ark:- ark,scp:ark:/home/jensen/Document/feats.ark,scp:/home/jensen/Document/feats.scp returned 255 ! Traceback (most recent call last): File "test.py", line 13, in kaldi_io.write_mat(f, v, k) File "/home/jensen/.local/lib/python3.8/site-packages/kaldi_io/kaldi_io.py", line 554, in write_mat fd.write(m.tobytes()) BrokenPipeError: [Errno 32] Broken pipe

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "test.py", line 13, in kaldi_io.write_mat(f, v, k) BrokenPipeError: [Errno 32] Broken pipe

[faber6911]

@jensen199105 I solved it by importing WriteHelper, in particular using "from kaldiio import WriteHelper" and using a script like this:

abs_path = args.abs_path
ark_path_train = os.path.join(abs_path, 'data/train/train.ark')
scp_path_train = os.path.join(abs_path, 'data/train/train.scp')
ark_path_test = os.path.join(abs_path, 'data/test/test.ark')
scp_path_test = os.path.join(abs_path, 'data/test/test.scp')

start = time.time()

if not os.path.isfile(ark_path_train):

    writer = WriteHelper('ark,scp:{},{}'.format(ark_path_train, scp_path_train), compression_method=compression_method)

    noise_choice = {'music':659, 'noise':929, 'speech':425}

    for count, line in enumerate(open('../data/train/wav.scp')):
        # clean audio path
        utt, path = line.rstrip().split()
        # clean audio file
        clean_audio, _ = librosa.load(path, sr = sample_rate)
        # now for every noise type we augment n times the clean audio file using random noise audio files
        for noise_type in noise_choice:
            for aug in range(train_augmentation):
                noise_track = np.random.randint(0, noise_choice[noise_type])
                _, noise_path = open('../data/musan_{}.scp'.format(noise_type)).readlines()[noise_track].rstrip().split()
                noise_audio, _ = librosa.load(noise_path, sr = sample_rate)
                noisy_audio = add_noise(clean_audio, noise_audio, snr=random.choice([2.5, 7.5, 12.5, 17.5]))
                # write ark and associated scp file in train directory
                writer(utt,np.concatenate((clean_audio.reshape(1, -1), noisy_audio.reshape(1, -1))))

Using this strategy you will be able to create the ark file and the associated scp file.