The files that I need to use are encoded as latin-1 and the detectEncoding function does not support this. I suggest a new parameter for TextGrid: encoding=None, which when passed supersedes the detectEncoding function.

hey man, just an FYI: mlf format allows to segments to start/end at the same time, but that drives praat crazy. so

7400000 8100000 EY1
8100000 8100000 sp sp
8100000 10700000 IH1

just happens to make a "bad" textgrid unless you change those 8100000's slightly.

Praat text files can be either ASCII or UTF-16. Right now, textgrid.py's reading and writing functions don't handle UTF-16, however. To be made to do this, they have to recognize UTF-16 (or just non-ASCII, I'd suggest, by comparing characters against \x7F) when reading, and the class instances have to be annotated as to whether they contain any UTF-16 marks, a feature which must percolate upward to Tier and Grid.

I'm not sure what we want to do here, but since we're redoing the write functions anyways, it seems worthwhile to consider.

Currently, it is possible to use textgrid.py to create IntervalTiers which don't explicitly mention empty (null "mark") intervals. These are readable by Praat, but which behave weirdly.

The appropriate behavior seems to be as follows. Any time an IntervalTier is being written (either IntervalTier.write or TextGrid.write when the TextGrid contains an IntervalTier), the written text file needs to explicitly include empty intervals. However, this should not change the state of the IntervalTier instance itself. (This does suggest an interesting possibility: maybe the "read" methods should ignore empty intervals. Something to think about.) This should be relatively easy to do without any additional instance variables, and just requires edits to IntervalTier.write and TextGrid.write.

I've assigned this to myself, for the moment. It's just one of those "while True:" problems I think.

class IntervalTier(object)
def intervalContaining(self, time):
i = self.indexContaining(time)
if i :
return self.intervals[i]

indexContaining return 0， ‘ if 0 ’ return false ，then return None。 but must return intervals[0]

Keelan Evanini and Bob Lannon both tell me that HVite in HTK may produce MLF files with rounding errors in their timestamps. There is an undocumented fix in the Penn Forced Aligner, but it doesn't generalize to arbitrary sample rates, which is a requirement I need for my aligner project. I also don't know if this is true of all versions of HTK. I'll need some long MLF files to figure this all out, but it's something to take care of in the future. I'm assigning this one to myself.

While rounding may be useful for some applications, it harms others.

Consider audio at 22050 Hz sampling rate with length 74751 samples. With a hopsize of 256 samples and padding such that each frame is centered at hopsize / 2, 3 * hopsize / 2, 5 * hopsize / 2, etc., this should yield 291 frames. With the current default rounding, it produces 292 frames and is misaligned with other features I use. Because rounding cannot be turned off, I cannot fix this while using this library in its current state.

I think rounding should be optional. I also think that it should not be the default system behavior.

I'm getting overlapping IntervalTier segments and trying to find the cause for this.

There is a line in MLF.add that rounds a float and the divides it by the samplerate:

pmin = round(float(line[0]), round_digits) / samplerate

However, this seems to defy the purpose of rounding. Here's an example:

>>> round(1234.56789, 3)
1234.568
>>> round(1234.56789, 3) / 10000
0.12345679999999999

Maybe it's better to define by samplerate and round afterwards, like this:

pmin = round(float(line[0]) / samplerate, round_digits)

If you agree about this, I'm happy to file a pull request.

The TextGrid, IntervalTier, and PointTier classes have __min__ and __max__ methods defined to return the minTime and maxTime attributes. These two methods aren't included as Python magic methods, so you have to call them by name:

it = textgrid.IntervalTier()
it.add(5.0, 6.0, 'a segment')

it.__min__()

which returns the minTime, 0.0

If you call min(it), the IntervalTier returns the first interval Interval(5.0, 6.0, a segment), and min() with a TextGrid returns the first/top tier. Go ahead and close if this is all intended behavior. I just saw the double underscores and went for min() but then didn't get what I expected.

I just ran into some weird behavior in Praat when editing an interval tier and think I traced it back to an interval with duration 0 (at the place where no boundary is displayed in the screenshot); would it make sense to disallow Intervals with duration 0? Could be done by modifying the check at https://github.com/kylebgorman/textgrid/blob/master/textgrid/textgrid.py#L173.

See test_multiline_with_double_quotes in tests.py added in #9 which demonstrates the issue. Text is cut off because the regex matches a quote character too early

Hi, older versions used to support PointProcess, which is needed to read files that contain events such as GCI marks. Was there a specific reason that lead to removal? And/or: Any chance of re-inclusion?

The module should be uploaded to PyPi, with whatever changes are necessary to do so.

Would it be possible to make a pypi release? I keep having to install from master because of some issues with the current release, and things seem pretty stable on GitHub.

Hi,

Thanks a lot for coding this. I'm just a bit confused on how to use it. I've how a .mlf file from HTK that I need to get into TextGrid format.

Do I need to import textgrid into Python and then call some functions on it?

Or is there a command line that I can use? python textgrid.py filename.mlf ??

Some documentation would help.

Cheers

Right now, the intervalContaining function has an error where it returns None even if the corresponding interval exists, if that interval is the first index. This is because if i: evaluates to false if i is 0.

   def intervalContaining(self, time):
        """
        Returns the interval containing the given time point, or None if
        the time point is outside the bounds of this tier. The argument
        can be a numeric type, or a Point object.
        """
        i = self.indexContaining(time)
        if i: #This evaluates to false if i is None or if i is = 0 
             return self.intervals[i]