[normalizer] Module Configuration

The normalizer module removes undesirable acoustic characteristics from the audio, such as speaker and audio channel effects. Audio normalization is a preprocessing requirement in most audio analysis operations, such as speech to text.

Input and Output

Mode Input Output Description
BLOCK f nf In BLOCK mode, the normalization is calculated over the whole utterance. Recommended for short utterances.
RUNNING f nf In RUNNING mode, the normalization process continuously adapts to the acoustic properties of the input audio.
RUNNING f, w nf

If the normalizer module in RUNNING mode is supplied with a word label stream in addition to feature frames, the normalization process continuously adapts to areas of audio categorized as speech.

Example:

nf <- normalizer(_, f)
nf <- normalizer(BLOCK, f)
nf <- normalizer(_, f, w)

Parameters

Some of these parameters do not have an internal default value. You must either set the configuration parameter to a suitable value, or set the value in the corresponding action parameter.


_FT_HTML5_bannerTitle.htm