kahrendt/microwakeword

This dataset contains spectrogram features in an mmap ninja format intended to use for microWakeWord training. The features are generated using TensorFlow's microfrontend with the following settings:

            sample_rate=16000,
            window_size=30,
            window_step=20,
            num_channels=40,
            upper_band_limit=7500,
            lower_band_limit=125,
            enable_pcan=True,
            min_signal_remaining=0.05,
            out_scale=1,
            out_type=tf.uint16,

These features are not scaled or converted to float. To do so, multiply by a factor of 0.0390625 after casting to a float. The current version (June 8th, 2024) of microWakeWord doesn't automatically do this, but it will be implented.

The dinner_party_background file contains features from the CHiME6 training set to use while training, the CHiME6 dev and evaluation sets for validating ambient background, and all DipCo audios for testing ambient background.

The no_speech_background file contains features from the FMA-medium, FSD50K, and WHAM datasets for training. Any source audio clips less than 6 seconds long were repeated until at least that length. All spectrograms were split over 5 second non-overlapping intervals. The first 25 features were discarded.

The speech_background file contains features from the LibriSpeech training other and VOiCES datasets for training. Any source audio clips less than 6 seconds long were repeated until at least that length. All spectrograms were split over 5 second non-overlapping intervals. The first 25 features were discarded.

---

license: cc-by-nc-4.0