PRAAT (‘doing Phonetics by computer’) is a speech analysis software developed by Paul Boersma and David Weenink at the Institute of Phonetic Sciences at the University of Amsterdam. It is a shareware program, downloadable for free use from the PRAAT homepage (www.praat.org or http://www.fon.hum.uva.nl/praat/), running under, among others, the most common operating systems – Windows, Macintosh and Linux. The software is regularly up-dated;
PRAAT is a highly sophisticated tool for speech analysis with manifold functions in a simple, user-friendly layout. To mention just a few of its basic operations, with PRAAT a speech signal or a ‘speech object’ can be:
- read from the hard disk (or other drives) if recorded elsewhere
- handled practically (even relatively long sound files)
- segmented, labelled and transcribed, with the waveform and its duration (as well as, optionally, the spectrogram with formant and pitch contours) displayed in the same window as the text being typed.
Since the Help tutorial in PRAAT itself (‘Introduction to Praat’), the Praat beginners’ manual and the Praat short tutorial (in pdf format) are all available on-line on the PRAAT homepage and offer a thorough and complete introduction to the program in English, it seems unnecessary to provide a proper introduction here .
In order to transcribe with PRAAT you need to be familiar with the following procedures (a detailed description of these steps can be found in French in Delais-Roussarie, Durand, Lyche, Meqqori & Tarrier (2002) :
- how to create a sound file in .WAV form
- how to open a long sound file
- how to segment and label a speech waveform (creation of ‘textgrids’)
- how to duplicate, copy, rename and remove tiers
- how to edit, listen and transcribe a speech segment
- how to save a Textgrid.
Among the most obvious advantages of working with PRAAT are the following:
- you can listen and transcribe simultaneously
- the text is aligned with the corresponding audio sequence
- you can listen to the sequence you transcribe as many times as you wish, and for portions as short as you like
- you can locate any given sequence in the text (from single words to sentences)
- you can move easily within the speech signal by zooming in and out of the sequence
- you can directly extract sounds from the passage transcribed.
And if you wish to do more advanced tasks you can also extract labelled segments (e.g. all segments labelled ‘voiceless stop’) and perform a number of sophisticated phonetic and acoustic measurements, including durational, spectrographic, intensity, formant and pitch analyses.
Transpraat (a program developed by A. Meqqori for the French PFC project) converts the coded text files (or ‘textgrids’) created with PRAAT into ‘normal’ text files that you can edit, format and print with your customary word processing tool, such as Microsoft Word. Note, however, that the document will no longer be linked to the original audio signal and PRAAT textgrid. Thus all eventual corrections or modifications in the transcription should be done and recorded in the original textgrid in PRAAT.
In a PRAAT textgrid the transcriptions appear within PRAAT-specific codes as illustrated in the following 34 second sequence from a conversation below. In this ‘raw’ textgrid format, the codes link each ‘interval’ to the original sound file:
xmin = 1177.5648270086172
xmax = 1181.1347845804989
text = "F: And how did you find Afghanistan, I mean the country?"
xmin = 1181.1347845804989
xmax = 1191.169782935204
text = "RM: Er, it's, it's, er, yeah, it was quite a nice place er, (XX) smelly in some places, the (XX) particularly, er it's very run down and er, "
xmin = 1191.169782935204
xmax = 1196.0005442176871
text = "RM: (stammering) most of the places, like the, the mountains or the farmers and stuff, they're, "
xmin = 1196.0005442176871
xmax = 1202.334784580499
text = "RM: they're generally quite friendly but er, very, very poor, it's like er, stuck in er, the Middle Ages I think."
xmin = 1202.334784580499
xmax = 1211.4604081632654
text = "RM: But er, yeah, it was not bad, (XX) not bad. Very rocky, the mountains, some of the, the mountains we were up were p/ particularly high."
The same sequence can be seen below, converted by Transpraat: there are no codes and speaking turns are automatically adjusted.
F: And how did you find Afghanistan, I mean the country?
RM: Er, it's, it's, er, yeah, it was quite a nice place er, (XX) smelly in some places, the (XX) particularly, er it's very run down and er, (stammering) most of the places, like the, the mountains or the farmers and stuff, they're, they're generally quite friendly but er, very, very poor, it's like er, stuck in er, the Middle Ages I think. But er, yeah, it was not bad, (XX) not bad. Very rocky, the mountains, some of the, the mountains we were up were p/ particularly high.
Dolmen is a free, open-source cross-platform application for the analysis of linguistic data. It was designed and is regularly updated by Julien Eychenne for Windows, Mac OS X and Linux.
"It offers a user-friendly interface to manage, annotate and query language corpora. It is particularly well suited for dealing with time-aligned data. The main features it offers are:
- Project management: organize files into projects and manage versions.
- Extensible metadata: files can be annotated with properties, which allow you to sort and organize your data.
- Interaction with Praat: Dolmen can read TextGrid files and open files directly in Praat.
- Powerful search engine: build and save complex queries; search patterns across tiers. • Standard-based: Dolmen files are encoded in XML and Unicode.
- Dolmen runs on all major platforms (Windows, Mac OS X and GNU/Linux) and is freely available under the terms of the GNU General Public License (GPL)."
(see http://julieneychenne.info/files/dolmen_manual.pdf for further information on the toolbox ; in particular check 6.5. if you are a PAC or a PFC user)
Dolmen is downloadable at www.julieneychenne.info/dolmen