Methods for scoring forced alignments...currently in development

A.TextGrid

File type = "ooTextFile"
Object class = "TextGrid"

xmin = 0
xmax = 3
tiers? <exists>
size = 1
item []:
    item [1]:
        class = "IntervalTier"
        name = "word"
        xmin = 0
        xmax = 3
        intervals: size = 5
        intervals [1]:
            xmin = 0
            xmax = 0.46004543779933693
            text = ""
        intervals [2]:
            xmin = 0.46004543779933693
            xmax = 1.1252957550452332
            text = "THAN"
        intervals [3]:
            xmin = 1.1252957550452332
            xmax = 1.966877481681608
            text = "I"
        intervals [4]:
            xmin = 1.966877481681608
            xmax = 2.5920524786114862
            text = "DID"
        intervals [5]:
            xmin = 2.5920524786114862
            xmax = 3
            text = ""

alignment.py

#!/usr/bin/env python
#
# Copyright (c) 2011 Kyle Gorman
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
# THE SOFTWARE.
# 
# alignment.py: Code for scoring hypothesized alignments against a reference
# alignment, using techniques from text segmentation resarch
# 
# Kyle Gorman <kgorman@ling.upenn.edu>
# 
# This code was developed to evaluate Prosodylab-Aligner, available at:
# 
# http://prosodylab.org/tools/aligner/

from textgrid import TextGridFromFile # https://github.com/kylebgorman/textgrid.py


## SCORING METHODS

def Pmu(ref, hyp, dt=0.001):
    """
    P_mu segmentation score, based loosely on:

    D. Beeferman, A. Berger, J. Lafferty. 1997. Text segmentation using 
    exponential models. Proceedings of EMNLP. 

    It is, to a first approximation, the probability that a non-null segment is
    misaligned. 

    >>> ref = TextGridFromFile('A.TextGrid')[0]
    >>> print round(Pmu(ref, TextGridFromFile('B.TextGrid')[0]), 3)
    0.713
    >>> print round(Pmu(ref, TextGridFromFile('C.TextGrid')[0]), 3)
    0.513
    """
    c = 0
    d = 0
    t = ref[0].minTime
    stop = ref[-1].maxTime - dt
    while t <= stop:
        ri = ref.intervalContaining(t)
        if ri and ri.mark == hyp.intervalContaining(t).mark: 
            c += 1
        else: 
            d += 1
        t += dt
    return c / float(c + d)


def Pk(ref, hyp, dt=0.001):
    """
    P_k segmentation score, based on:

    D. Beeferman, A. Berger, J. Lafferty. 1999. Statistical models of text
    segmentation. Machine Learning 34(1-3): 177-210.

    It is, to a first approximation, the probability that some span in the 
    hypothesized segmentation crosses a segment boundary iff it crosses a 
    segment boundary in the reference segmentation.

    >>> ref = TextGridFromFile('A.TextGrid')[0]
    >>> print round(Pk(ref, TextGridFromFile('B.TextGrid')[0]), 3)
    0.63
    >>> print round(Pk(ref, TextGridFromFile('C.TextGrid')[0]), 3)
    0.404
    """
    c = 0
    d = 0
    k = (x.maxTime - x.MinTime for x in ref) / len(ref) / 2.
    t = ref[0].minTime
    stop = ref[-1].maxTime - k
    while t <= stop:
        rs = ref.intervalContaining(t) == ref.intervalContaining(t + k)
        hs = hyp.intervalContaining(t) == hyp.intervalContaining(t + k)
        if rs == hs: 
            c += 1
        else: 
            d += 1
        t += dt # increment
    return c / float(c + d)


def Ddiff(ref, hyp, dt=0.01):
    """
    1 - WindowDiff segmentation score, based on:

    L. Pevzner, M. A. Hearst. 2002. A critique and improvement of an evaluation
    metric for text segmentation. Computational Linguisics 28(1): 19-36.

    >>> ref = TextGridFromFile('A.TextGrid')[0]
    >>> print round(Ddiff(ref, TextGridFromFile('B.TextGrid')[0]), 3)
    0.631
    >>> print round(Ddiff(ref, TextGridFromFile('C.TextGrid')[0]), 3)
    0.402
    """
    c = 0
    d = 0
    k = mean([x.maxTime - x.minTime for x in ref]) / 2.
    t = ref[0].minTime
    stop = ref[-1].maxTime - k
    while t <= stop:
        rb = ref.indexContaining(t + k) - ref.indexContaining(t)
        hb = hyp.indexContaining(t + k) - hyp.indexContaining(t)
        d += abs(rb - hb) > 0
        c += 1
        t += dt
    return float(c - d) / c


if __name__ == '__main__':
    import doctest
    doctest.testmod()

B.TextGrid

File type = "ooTextFile"
Object class = "TextGrid"

xmin = 0
xmax = 3
tiers? <exists>
size = 1
item []:
    item [1]:
        class = "IntervalTier"
        name = "word"
        xmin = 0
        xmax = 3
        intervals: size = 5
        intervals [1]:
            xmin = 0
            xmax = 0.36004543779933693
            text = ""
        intervals [2]:
            xmin = 0.36004543779933693
            xmax = 1.3252957550452332
            text = "THAN"
        intervals [3]:
            xmin = 1.3252957550452332
            xmax = 1.866877481681608
            text = "I"
        intervals [4]:
            xmin = 1.866877481681608
            xmax = 2.6920524786114862
            text = "DID"
        intervals [5]:
            xmin = 2.6920524786114862
            xmax = 3
            text = ""

C.TextGrid

File type = "ooTextFile"
Object class = "TextGrid"

xmin = 0
xmax = 3
tiers? <exists>
size = 1
item []:
    item [1]:
        class = "IntervalTier"
        name = "word"
        xmin = 0
        xmax = 3
        intervals: size = 5
        intervals [1]:
            xmin = 0
            xmax = 0.06004543779933693
            text = ""
        intervals [2]:
            xmin = 0.06004543779933693
            xmax = 1.6252957550452332
            text = "THAN"
        intervals [3]:
            xmin = 1.6252957550452332
            xmax = 1.766877481681608
            text = "I"
        intervals [4]:
            xmin = 1.766877481681608
            xmax = 2.8920524786114862
            text = "DID"
        intervals [5]:
            xmin = 2.8920524786114862
            xmax = 3
            text = ""