Ilyushin / signal-transformation

Widely used signal transformation using TensorFlow API.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Signal transformation

PyPI version

Helps to transform signal data like speech into different representations using TensorFlow 2 API.

pip3 install signal-transformation

Mantains following types of output formats of the voice:

  • PCM
  • STFT
  • MAGNITUDE
  • MFCC
  • LOG MFCC

Voice

  1. Transform wav files to needed format and then save them to tf_records

    import pandas as pd
    from signal_transformation.voice.tftr import wav_to_tf_records, SpectFormat    
    
    metadata_path = '/path/to/parquet/file'
    output_dir = '/output/directory'
    
    metadata = pd.read_parquet(metadata_path)
    
    wav_to_tf_records(
        metadata=metadata,
        output_dir=output_dir,
        spec_format=SpectFormat.MAGNITUDE
    )    

    Or use as script

    wav_to_tf_records --metadata_path="./vox1_metadata.gzip" --spect_format=3 --sample_rate=16000 --num_mfcc=13 --spect_shape="300, 200, 1" --num_shards=512 --output_dir="/dataset/tf_records"
    
  2. Example of creating a metadata file of VoxCeleb 1

    from signal_transformation import helpers   
    
    helpers.voxceleb_files_to_metadata('files_dir', 'dev', 'vox1_metadata.gzip')

    Or use as a script

    vox_to_metadata --files_dir="/vox2/audio/dev/wav/" --part_name="dev" --output_file="/vox2/audio/dev/vox2_wav_metadata.gzip"
    
  3. Create overlapping signals dataset

    from signal_transformation import helpers
    
    
    input_folder = '/path/to/wav/files'
    output_folder = '/output/directory'
    
    result = helpers.prepare_data(
        input_folder,
        output_folder,
        size=100,
        speakers_number=2,
        overlapping=True
    )    

Images

  1. Imagenet dataset to TensorFlow records
imagenet_to_tf_records --mode="grayscale" --train_data_dirs "/datasets/images/imagenet/train_1/raw /datasets/images/imagenet/train_2/raw" --validation_data_dir /datasets/images/imagenet/validation --validation_labels_file /datasets/images/imagenet/ILSVRC2012_devkit_t12/data/ILSVRC2012_validation_ground_truth.txt --output_dir /datasets/images/imagenet/tf_records 

About

Widely used signal transformation using TensorFlow API.

License:MIT License


Languages

Language:Python 99.4%Language:Shell 0.6%