Module CuedSpeech.whenhand.transition

Class WhenTransitionPredictorAttina

Description

Predict hand transition moments with (Attina, 2005) method.

Virginie Attina Dubesset (2005) La langue française parlée complétée (LPC) : production et perception. PhD Thesis of INPG Grenoble, France. Page 117, page 136 and page 148 https://tel.archives-ouvertes.fr/file/index/docid/384080/filename/these_attina.pdf

See page 112 (Experiment 1, no shape change, position changes, CV syllables):

The average key duration is 399.5ms
M1A1: mean duration is 239 ms (std is 87ms) => 60%
A1M2: mean duration is 37 ms (std is 76ms) => 9%

See page 113 (Experiment 1, no shape change, position changes, V syllables):

The average key duration is 383.6ms
M1A1: mean duration is 183 ms (std is 79ms) => 47%
A1M2: mean duration is 84 ms (std is 64ms) => 22%

Both results are summarized in the scheme page 117.

See page 118 (Experiment 2, shape changes, position doesn't), results page 123-124:

The average CV key duration is 275ms
D1A1: mean duration is 124 ms (std is 34ms)
A1D2: mean duration is 46 ms (std is 35ms)

See page 119 (Experiment 2, both shape/position change), results page 124-

The average CV key duration is 316.3ms
D1A1: mean duration is 171 ms (std is 48ms) => 54%
A1D2: mean duration is -3 ms (std is 45ms) => 0%
M1A1: mean duration is 205 ms (std is 54.5ms) => 65%
A1M2: mean duration is 33 ms (std is 50ms) => 10%

Both results are summarized in the scheme page 128.

A general synchronization scheme is proposed page 136 with percentages instead of durations; and extended to 3 more speakers page 148.

The percentages implemented in this class are related to an average key duration of 399.5ms (Experiment 1).

Constructor

Instantiate (Attina, 2005) hand transition moment's predictor.

View Source

def __init__(self):
    """Instantiate (Attina, 2005) hand transition moment's predictor.

    """
    super(WhenTransitionPredictorAttina, self).__init__()
    self._description = MSG_DESCRIPTION_ATTINA
    self.set_static_duration(0.4)

Public functions

predict_position

Predict [M1,M2] the moments when the hand is moving.

To predict M2: - if CV key: both page 136 and page 117, M2 is 10% later than A1. - if -V key: page 117, M2 is 21% later than A1. - if C- key: no information. CV key case is used.

To predict M1: - if CV key: page 136 M1 is 62-65% of A1A3, but page 117, M1 is 60% of A1A3 - if -V key: page 117, M1 is 46% of A1A3 - if C- key: no information. CV key case is used.

Parameters

isnilshape: (bool) True is the key is of the form "-V", i.e. no consonant

Returns

tuple(m1: float, m2: float)

View Source

def predict_position(self, is_nil_shape: bool=False, **kwargs) -> tuple:
    """Predict [M1,M2] the moments when the hand is moving.

        To predict M2:
            - if CV key: both page 136 and page 117, M2 is 10% later than A1.
            - if -V key: page 117, M2 is 21% later than A1.
            - if C- key: no information. CV key case is used.

        To predict M1:
            - if CV key: page 136 M1 is 62-65% of A1A3, but page 117, M1 is 60% of A1A3
            - if -V key: page 117, M1 is 46% of A1A3
            - if C- key: no information. CV key case is used.

        :param is_nil_shape: (bool) True is the key is of the form "-V", i.e. no consonant
        :return: tuple(m1: float, m2: float)

        """
    a3a1 = self.get_a1a3_avg_duration()
    if is_nil_shape is False:
        a1m2 = a3a1 * 0.1
        m1a1 = a3a1 * 0.62
    else:
        a1m2 = a3a1 * 0.21
        m1a1 = a3a1 * 0.46
    m1 = max(0.0, self.a1 - m1a1)
    m2 = self.a1 + a1m2
    return (m1, m2)

predict_shape

Predict [D1,D2] the moments when fingers are changing.

D1 and D2 are predicted from the key duration only. Following the model proposed page 136:

D2 is right before A1 (1%).
Estimating D1D2 interval, which is 55% of A3A1, gives D1 position related to D2.

Returns

(d1: float, d2: float)

Parameters

isnilshape

View Source

def predict_shape(self, is_nil_shape: bool=False, **kwargs) -> tuple:
    """Predict [D1,D2] the moments when fingers are changing.

        D1 and D2 are predicted from the key duration only. Following the
        model proposed page 136:

        - D2 is right before A1 (1%).
        - Estimating D1D2 interval, which is 55% of A3A1, gives D1 position
          related to D2.

        :return: (d1: float, d2: float)

        """
    a3a1 = self.get_a1a3_avg_duration()
    if is_nil_shape is False:
        d2d1 = a3a1 * 0.54
    else:
        d2d1 = a3a1 * 0.36
    d1 = max(0.0, self.a1 - d2d1)
    d2 = self.a1
    return (d1, d2)