SPPAS
Automatic Phonetic Annotation of Speech
Overview
Project by Laboratoire Parole et Langage, Aix-en-Provence, France.
Description:
SPPAS is a tool to produce automatically phonetic annotations from a recorded speech sound and its transcription. The whole procedure is a succession of automatic steps. Result is a set of TextGrid files. SPPAS is open source software issued under the GNU Public License.
SPPAS is currently designed for French, English, Italian and Chinese and there is an easy way to add other languages.
Operating systems are: Linux, MacOS and Windows.
Screenshots of the version 1.4.2:
|
|
|---|
Screenshots of the version 1.4:
|
|
|---|---|
|
|
SPPAS over a demonstration...
An input example
The input file is the following word-tokenized ASCII text:
they only played two movies which we had both already seen
I never get to sleep on the airplane because it's so uncomfortable
This is the oriana1.txt file. It is located in the samples/samples-EN directory included in the SPPAS package. The corresponding wav file is oriana1.wav in the same directory.
Momel/INTSINT
If a file oriana1.hz or oriana1.PitchTier exists in the same directory as oriana1.wav, the Momel and INTSINT annotations can be activated. It produces a oriana1.momel.TextGrid file with 2 point tiers.
IPUs Segmentation
The Inter-Pausal Units segmentation takes as input the oriana1.txt and the oriana1.wav files.
The output oriana1.TextGrid file contains one tier named IPU. This segmentation depends on the number of silences indicated in the transcription by newlines and/or the '#' symbol.
Phonetization
The phonetization takes as input the oriana1.TextGrid file. This phonetization is performed only on the basis of a dictionary that can be manually changed.
It creates the oriana1-phon.TextGrid file and oriana1-tokens.TextGrid. By convention, spaces separate words, dots separate phones and pipes separate phonetic variants of a word. For example, the sentence I never get to sleep on the airplane will be phonetized as: ay n.eh.v.er g.eh.t|g.ih.t t.uw|t.ix|t.ax s.l.iy.p aa.n|ao.n dh.ax|dh.ah|dh.iy eh.r.p.l.ey.n
Alignment
The alignment performs a phonetic segmentation from the oriana1-phon.TextGrid, oriana1-tokens.TextGrid and oriana1.wav files. Alignment is performed on the basis of an acoustic model and on both phonetization and transcription files. The Julius speech recognition engine is used to estimate the temporal boundaries of each word, its phonetization and the temporal boundaries of each phoneme of that word.
It creates the oriana1-phon.palign.TextGrid file which includes the phoneme alignments, and the word alignments.
Syllabification
The last automatic step is syllabification. This step is available only for French and Italian.
This syllabification is performed on the basis of the phoneme alignments with a rule-based system. Two main principles are followed: 1. there is one and only one vowel per syllable, and 2. pauses are syllabic boundaries. Rules are fixed to split the consonant clusters.
SPPAS outputs:
This is the list of output files:
- a TextGrid file with Momel/INTSINT annotation
- a TextGrid file with utterance segmentation
- two TextGrid files with phonemes and word segmentations
- a TextGrid file with syllable segmentation
This is the final result, with TextGrid files merged in Praat:
