PART 1

SPPAS: why, what for, and for whom?

Getting to Know SPPAS

SPPAS: The automatic annotation and analysis of speech

• Distributed since 2012, primarily for speech segmentation:
  • 5 annotations
  • 4 languages

• Version 4.25 (May 2025):
  • 23 annotations
  • 15 languages
  • annotation analyses
  • file conversion

SPPAS-4.25 in a few numbers

30, 170, 550, 1400, 110k, 90k and 1.

Open Science award - 2022

Getting SPPAS

https://sppas.org/

https://sourceforge.net/projects/sppas/

• SPPAS is registered with the program protection agency under the reference: IDDN.FR.001.500008.000.S.C.2024.000.31235

SPPAS Citations

SPPAS is only cited for its original phonetic segmentation function
... and almost exclusively by the initially targeted users!

Except one: SPPAS was used for LipSync in Final Fantasy VII Remake

Getting to Know Speech Segmentation

Speech Segmentation

the process of taking the orthographic transcription text of an audio speech segment, like IPUs^*, and determining where particular phonemes/words occur in this speech segment.

In 2011... Speech Segmentation

• could be done automatically with toolkits such as:
  • HTK
  • Julius
  • Sphinx
  • …
• Linguists do it (almost) manually.

Reality Check: Two Disciplines, Two Practices

It's only one purpose but... multiple needs!

• Phonetic segmentation for:
  1. Conversational speech in French
  2. From patients (Parkinson's disease, ALS, …), in French
  3. Collected during psycholinguistic experiments, in English
  4. For prosodic analyses, in Italian, with words and non-words
  5. Of read texts, in Mandarin
• … and many other needs, including a user-friendly interface (obviously!)

      

Reality Check: Two Disciplines, Two Practices

Getting SPPAS: Your Data, Your Way

SPPAS allows users to adapt the software to their needs

Theoretical positioning: consider methods as language- and task-independent as possible.

• language-independent algorithms: only language-dependent resources
• ... easy and fast to add a new language
• ... easy and fast to modify existing resources

• Main consequences:
  1. the possibility of adapting technologies to low-resource languages
  2. the possibility of adapting them to a specific task

Solution example:
Resources can be modified

SPPAS resources are under open licenses and in readable file formats.

Example of open and editable resource:

Solution example:
Adapting to a New Language is 4 Resources to Create

• By a linguist:
  • a word list for the language
  • a pronunciation dictionary for the language
  • a few minutes (around 5) of manually segmented speech
• By me (or another computer scientist):
  • the acoustic model, trained from the few minutes of segmented speech

All of this in a .zip file, writing the documentation, and that’s it!

Solution example:
a specific item for the hesitation ("fp")

14 "euh" in 18 seconds...
In this corpus, they represent 6% of the tokens.

Solution example:
a specific item for laughter

9 laughs in 19 seconds...
In this corpus, they represent 4% of the tokens.

Getting Ready for SPPAS: Preparing Your Data

Step 1: Recording speech

• One channel per speaker
• Anechoic room, or an environment with no/low noise
• Audio, for automatic annotation tools:
  • Any uncompressed file format, commonly WAVE
  • 16000Hz / 16bits is enough
• Audio, for manual annotation tools:
  • Any un-compressed file format
  • 32000-48000-96000Hz / 16bits

Step 2: Search for Inter-Pausal Units

• Parameters to define manually:
  • fix the minimum silence duration
  • fix the minimum speech duration
• As results:
  • speech and silences are time-aligned automatically
  • a manual verification is highly recommended

Step 3: Orthographic Transcription

• Any transcription must follow a convention
• In speech (particularly in spontaneous speech), many phonetic variations occur:
  • some of these phonologically known variants are predictable
  • but many others are still unpredictable (especially invented words, regional words or words borrowed from another language)
• The orthographic transcription must be enriched:
  • it must be a representation of what is “perceived” in the signal.

Step 3: Enriched Orthographic Transcription

A transcription convention that allows SPPAS to process speech events.

Result example

Getting the Data Extract for Practice

Cheese! Corpus

• 11 interactions (mixed & single-gender), ~15 min each
  • Dyad partners are friends outside university
  • Aix-Marseille Univ, Language Sciences students (L2–PhD)
  • Native French speakers
• Audio & video in anechoic booth
  • Face-to-face seating
  • Headset microphones for each speaker / Individual camera

Step 1: MA-PC Dialog Extract

Step 2: Search for IPUs

Step 3: IPUs Manually corrected and Transcribed (with Praat)

Step 3: IPUs Manually corrected and Transcribed with Praat

Getting started with SPPAS

Communicate with SPPAS

1. The powerful solution is with Python and its API:
   • See https://sppas.org/scripting.html
2. Another reproducible solution is with the embedded Python programs:
   • See sppas/bin and sppas/scripts files
3. The most often used solution is its Graphical User Interface (GUI):
   • Double-click sppas.bat under Windows, sppas.command for macOS/Linux

SPPAS GUI: the Log window

SPPAS displays its messages here instead of opening multiple modal dialogs each time you perform an action!

A red or an orange message should alert you

SPPAS GUI: the Main window is a notebook

List of ALL Tasks to Perform any Automatic Annotation
Speech segmentation included

1. In the "Files" tab:

   • add the files to be annotated;
   • check them;
   
2. In the "Annotate" tab:

   • follows the steps;
   • click “Let’s go!”.
   
3. Wait and read the report (Yes, that’s really it-there's nothing more!)

STANDALONE Annotations Required for Speech Segmentation

Files after Speech Segmentation

PART 2

MarsaTag: syntactic analysis of spoken language

POS-Tagging

• It is mostly dedicated to written text, in the NLP community
• A system must be adapted to deal with speech, particularly for conversational speech:
  • spoken data are time-aligned, and we expect to get a time-aligned morpho-syntax!
  • the lexicon and the probabilities of tokens are different between written texts and speech, so they must be updated.

MarsaTag: a tool for French POS-Tagging

• Developed by Stéphane Rauzy, Research Engineer at CNRS’s Laboratoire Parole et Langage.

MarsaTag plugin for SPPAS

1. Download and install MarsaTag from Ortolang:
   https://www.ortolang.fr/market/tools/sldr000841

2. Download and install MarsaTag-Plugin
   https://sppas.org/download.html

• Data preparation: Perform Speech Segmentation to create '-palign' files

Use MarsaTag plugin

1. In the "Files" tab:

   • add the files to be annotated;
   • check them;
   
2. In the "Plugins" tab:

   • click the plugin logo and follow instructions
   • click “Let’s go!”.
   
3. Wait and read the report

MarsaTag annotation result

PART 3

Other phonetic and prosodic annotations

Momel and INTSINT

Proposed by Daniel Hirst, Directeur de Recherches Emeritus at CNRS's Laboratoire Parole et Langage.

Momel (modelling melody)

• algorithm modelling raw fundamental frequency curves with a quadratic spline function target F0 Points

INTSINT: an INternational Transcription System for INTonation

• based on an inventory of minimal pitch contrasts found in published descriptions of intonation patterns
• surface phonological structure
• mapping from Momel anchor points to INTSINT tones:
  • Absolute tones: T(op) M(id) B(ottom)
  • Relative tones: H(igher) S(ame) L(ower)
  • Iterative relative tones: U(pstepped) D(ownstepped)

Data preparation

• Estimate a F0 value every 10ms from the audio file
• Save into one of the following formats:
  • .hz file with one value per line (0 if no estimated F0 value)
  • .PitchTier file with Praat

Momel and INTSINT annotation process

1. In the "Files" tab:
   • add the files to be annotated;
   • check them;
2. In the "Annotate" tab:
   • follows the steps, including checking "Momel" and "INTSINT STANDALONE annotations;
   • click “Let’s go!”.
3. Wait and read the report

Momel annotation result

INTSINT annotation result

Syllables

Description

• The system is based on 2 main principles:
  1. a syllable contains a vowel, and only one;
  2. a pause is a syllable boundary.
• Fix rules to find the boundaries between two vowels

Syllables annotation process

• Data preparation: Perform Speech Segmentation to create '-palign' files

1. In the "Files" tab:
   • add the files to be annotated;
   • check them;
2. In the "Annotate" tab:
   • follows the steps, including checking "Syllabification" STANDALONE annotation;
   • click “Let’s go!”.
3. Wait and read the report

Syllables annotation result

Time Group Analyzer

Proposed by Dafydd Gibbon, Fakultät für Linguistik und Literaturwissenschaft, Universität Bielefeld, Germany.

Time Group Analyzer

Analysis of speech rhythm based on syllable durations

TGA annotation process

• Data preparation: perform Syllabification to create '-syll' files

1. In the "Files" tab:
   • add the files to be annotated;
   • check them;
2. In the "Annotate" tab:
   • follows the steps, including checking "TGA" STANDALONE annotation;
   • click “Let’s go!”.
3. Wait and read the report

TGA annotation result

The full SPPAS annotation workflow

PART 4

Other features

File conversion

SPPAS is interoperable by design

Convert files

1. In the "Files" tab:
   • add the files to be converted;
   • check them;
2. In the "Convert" tab:
   • Check the expected file format;
   • click “Perform the conversion”.
3. Wait and read the report at the bottom of the tab

Convert tab

Conversion result

Filtering annotated data

Filtering annotated data

• Efficient on any supported file format
• Allows to get a tier with selected annotations only
• Two types of filters:
  1. apply criteria on one tier
  2. apply time relations on two tiers

Filtering annotated data: example (1)

Filtering annotated data: example (2)

Annexes

References

• Bigi Brigitte (2015). SPPAS - MULTI-LINGUAL APPROACHES TO THE AUTOMATIC ANNOTATION OF SPEECH. The Phonetician. Journal of the International Society of Phonetic Sciences, Journal of ISPhS/International Society of Phonetic Sciences, vol. 111-112 (ISSN:0741-6164), pp.54-69.
• Priego-Valverde Béatrice, Brigitte Bigi, Salvatore Attardo, Lucy Pickering and Elisa Gironzetti (2018). “Is smiling during humor so obvious? A cross-cultural comparison of smiling behavior in humorous sequences in American English and French interactions”, Intercultural Pragmatics, 15(4): 563–591.
• Bigi Brigitte, Zimmermann Maryvonne, André Carine (2022). CLeLfPC: a Large Open Multi-Speaker Corpus of French Cued Speech. In Proceedings of The 13th Language Resources and Evaluation Conference, pp. 987-994, Marseille, France.
• Rauzy Stéphane, De Montcheuil Grégoire, Blache Philippe (2014). MarsaTag, a tagger for French written texts and speech transcriptions. Second Asia Pacific Corpus Linguistics Conference (2014 March 7-9 : Hong Kong). 2014.
• Hirst Daniel, Espesser Robert (1993). Automatic modelling of fundamental frequency using a quadratic spline function. Travaux de l’Institut de Phonétique d’Aix, 15:75–85.
• Gibbon Dafydd (2013). TGA: a web tool for Time Group Analysis, TRASP - Interspeech satellite event

About

• Document author: Brigitte Bigi
• Copyright : © 2025 Brigite Bigi, CNRS, LPL, Aix-en-Provence, France
• Last modification : 2025-06-10
• https://sppas.org/bigi/Doc/2025-ANF-ACORPO/

License

This document is a creative work, the exclusive property of LPL, protected by French and international intellectual property law, and licensed under CC BY-NC-ND (Attribution / Non-Commercial / No Derivatives). This license permits any distribution (sharing, copying, reproducing, distributing, communicating), except for commercial purposes, by any means and in any format, provided that the work is distributed without modification and in its entirety. You are free to copy, distribute, and transmit this document, provided that you credit the SPPAS project.