edu.cmu.sphinx.frontend.feature

Class BatchCMN

java.lang.Object

edu.cmu.sphinx.util.props.ConfigurableAdapter

edu.cmu.sphinx.frontend.BaseDataProcessor

edu.cmu.sphinx.frontend.feature.BatchCMN

All Implemented Interfaces:

DataProcessor, Configurable

public class BatchCMN
extends BaseDataProcessor

Applies cepstral mean normalization (CMN), sometimes called channel mean normalization, to incoming cepstral data. Its goal is to reduce the distortion caused by the transmission channel. The output is mean normalized cepstral data.

The CMN processing subtracts the mean from all the Data objects between a DataStartSignal and a DataEndSignal or between a SpeechStartSignal and a SpeechEndSignal. BatchCMN will read in all the Data objects, calculate the mean, and subtract this mean from all the Data objects. For a given utterance, it will only produce an output after reading all the incoming data for the utterance. As a result, this process can introduce a significant processing delay, which is acceptable for batch processing, but not for live mode. In the latter case, one should use the LiveCMN.

CMN is a technique used to reduce distortions that are introduced by the transfer function of the transmission channel (e.g., the microphone). Using a transmission channel to transmit the input speech translates to multiplying the spectrum of the input speech with the transfer function of the channel (the distortion). Since the cepstrum is the Fourier Transform of the log spectrum, the logarithm turns the multiplication into a summation. Averaging over time, the mean is an estimate of the channel, which remains roughly constant. The channel is thus removed from the cepstrum by subtracting the mean cepstral vector. Intuitively, the mean cepstral vector approximately describes the spectral characteristics of the transmission channel (e.g., microphone).

See Also:

LiveCMN

Field Summary

Fields inherited from class edu.cmu.sphinx.util.props.ConfigurableAdapter

logger

Constructor Summary

Constructors

Constructor and Description
BatchCMN()

Method Summary

Modifier and Type	Method and Description
Data	getData() Returns the next Data object, which is a normalized cepstrum.
void	initialize() Initializes this BatchCMN.
void	newProperties(PropertySheet ps) This method is called when this configurable component needs to be reconfigured.

Methods inherited from class edu.cmu.sphinx.frontend.BaseDataProcessor

getPredecessor, setPredecessor

Methods inherited from class edu.cmu.sphinx.util.props.ConfigurableAdapter

getName, initLogger, toString

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

BatchCMN

public BatchCMN()

Method Detail

newProperties

public void newProperties(PropertySheet ps)
                   throws PropertyException

Description copied from interface: Configurable

This method is called when this configurable component needs to be reconfigured.

Specified by:

newProperties in interface Configurable

Overrides:

newProperties in class ConfigurableAdapter

Parameters:

ps - a property sheet holding the new data

Throws:

PropertyException - if there is a problem with the properties.

initialize

public void initialize()

Initializes this BatchCMN.

Specified by:

initialize in interface DataProcessor

Overrides:

initialize in class BaseDataProcessor

getData

public Data getData()
             throws DataProcessingException

Returns the next Data object, which is a normalized cepstrum. Signal objects are returned unmodified.

Specified by:

getData in interface DataProcessor

Specified by:

getData in class BaseDataProcessor

Returns:

the next available Data object, returns null if no Data object is available

Throws:

DataProcessingException - if there is an error processing data