Last update, July, 2015

-

Overview

• SPPAS is a scientific computer software package written and maintained by Brigitte Bigi of the Laboratoire Parole et Langage, in Aix-en-Provence, France.

• Operating systems:

  -

• GNU Public License, version 3

Download, install and update

• Web site: http://sldr.org/sldr00800/preview/

  1. Follow carefully instructions of the installation page
  2. Download the last package
  3. Unzip on your computer

• Update SPPAS regularly:

  1. Put the old package into the Trash
  2. Download and unpack the new one

Speech files: recommendations

• only wav, aiff and au files

• channels: 1 (mono)

• sample width: 16 bits

• frame rates: 16000 Hz

• NEVER convert from a compressed file (mp3, ...)

• Good recording quality is expected

Open speech file(s) with SndRoamer component for a diagnosis

Annotated files: recommendations

• UTF-8 encoding only

• No accentuated characters in file names (nor in the path)

• Supported file formats to open/save (software, extension):

  • SPPAS: xra
  • Praat: TextGrid, PitchTier, IntensityTier
  • Elan: eaf
  • HTK: lab, mlf
  • Sclite: ctm, stm
  • Phonedit: mrk
  • Excel/OpenOffice/R/...: csv
  • subtitles: sub, srt
• Supported file formats to import/export (software, extension):
  • AnnotationPro: antx
• Supported file formats to import (software, extension):
  • Transcriber: trs
  • Anvil: anvil

By using SPPAS, you agree to cite references in your publications

• As for example:

Brigitte Bigi (2012). “SPPAS: a tool for the phonetic segmentations of Speech”, Language Resources and Evaluation Conference, ISBN 978-2-9517408-7-7, pages 1748-1755, Istanbul (Turkey).

Brigitte Bigi, Daniel Hirst (2012). “SPeech Phonetization Alignment and Syllabification (SPPAS): a tool for the automatic analysis of speech prosody”, Speech Prosody, Tongji University Press, ISBN 978-7-5608-4869-3, pages 19-22, Shanghai (China).

• Other references are available in the documentation, and related PDF files are included in the package.

What SPPAS can do?

• Automatic Annotations:

  • Momel/INTSINT: Modelling melody
  • IPUs segmentation: utterance level segmentation
  • Tokenization: text normalization
  • Phonetization: grapheme to phoneme conversion
  • Alignment: phonetic segmentation
  • Syllabification: group phonemes into syllables
  • Repetitions: detect self-repetitions

• ... and many other things!

  • Components
  • Plugins

Components and plugins

• IPUScribe: Manual transcription

• SndRoamer: Play sound (mono wav)

• Statistics: Estimates/Save statistics of tiers

• DataRoamer: Manipulate annotated files

• DataFilter: Select/Filter annotations of tiers

• SppasEdit: Display wav and annotated files

• TierMapping-plugin: Create tier by mapping annotations

• MarsaTag-plugin: Use the POS-Tagger MarsaTag from SPPAS (French only)

Usage: GUI, CLI or Python Scripts

• Read documentation for command-line interface and python scripts
• Graphical User Interface: -

GUI Usage (1)

• Open the file explorer of your system

• Go to the SPPAS folder location

• Windows:
  • Doucle-click on the sppas.bat file
• MacOS / Linux:
  • Double-click on the sppas.command file

GUI Usage (2)

• Click on the 'Add File' button

• Explore the samples folder and choose as many wav files as expected

• All files with the same name as the selected wav files will be added into the list

• Click (and/or ctrl+click) on some files in this list

• Choose what you want to do with your selection (a component, automatic annotations, plugin)

One of the specificy of SPPAS...

All the automatic annotations are based on language independent approaches

-

• This means:

  1. adding a new language consist in adding related resources (lexicons, dictionaries, etc)
  2. any user can edit resources to modify them to adapt automatic annotations to its own requirements

Phonetic Segmentation

• Definition:

The process of taking the text transcription of an audio speech segment and determining where in time particular phonemes occur in the speech segment

• Manual vs Automatic?

-

Automatic Speech segmentation: in 3 steps

-

Inputs: Orthographic Transcription / Speech signal

• Enriched Orthographic transcription:

  • Representation of what is “perceived” in the signal
  • Already time-aligned at the utterance level (IPUs segmentation)

  • It must includes:

    • Filled pauses
    • Short pauses
    • Repeats
    • Noises and Laugh

• Audio: mono wav file, 16KHz, 16 bits

Tokenization (automatic step 1)

• Tokenization requires a list of words (lexicon)

• To create/edit a lexicon:

  • create/open the file SPPAS/resources/vocab/LANG.vocab
  • save (UTF-8 encoding)

• Input example:

  Et euh donc donc du coup c'est toi c'est un peu toi q(ui) a les premiers contacts avec le avec le gosse quoi + et puis là ils te demandent le prénom donc faut ce soit prêt là @ parce que putain.

• Output:

  et euh donc donc du coup c' est toi c'est un_peu toi qui a les premiers contacts avec le avec le gosse quoi + et puis là ils te demandent le prénom donc faut ce soit prêt là @ parce_que putain

Phonetization (automatic step 2)

• Phonetization requires a pronunciation dictionary

• To create/edit a dictionary:

  • create/open the file SPPAS/resources/dict/LANG.dict
  • save (UTF-8 encoding)

• In the phonetization output, by convention, spaces separate words, dots separate phones and pipes separate phonetic variants of a word. Example:

  • input: the flight
  • output: dh.ax|dh.ah|dh.iy f.l.ay.t

• If a word is missing of the dictionary, SPPAS generates a pronunciation.

Alignment (automatic step 3)

• Alignment requires an acoustic model

-

Outputs/Results

• Each automatic annotation generates a file and...

  • A “merged” file is also created

• Open such file(s) in the SppasEdit component, or Praat, or Elan, ...

  -

• Save/Export any file into any format (XRA, TextGrid, EAF, CSV) with one of the 'Export' buttons

That's all!

• You are now ready to test SPPAS with the proposed set of samples...

• ... and do not forget to read the documentation: it contains most of the answers to your questions!

-