---
title: "A schematic overview of the Emu Speech Database Management system"
author: "Jonathan Harrington"
date: "WiSe 2023"
output: 
  bookdown::html_document2:
    number_sections: TRUE
    toc: true
    theme: flatly
    highlight: pygments
---

<style>
div.gray {background-color: #e8e8e8; border-radius: 5px; padding: 20px;}
body {font-size: 16pt;}
h1 {font-size: 24pt;}
h2 {font-size: 22pt;}
p.caption {font-size: 12pt; text-align: justify;}
code.sourceCode {font-size: 16pt;}
</style>

# In the beginning
The core idea behind Emu today is just the same as it was when it first launched in the mid-late 1980s at [CSTR, Edinburgh University](https://www.cstr.ed.ac.uk) and known as *APS* which stood for [Acoustic Phonetics in S](https://www.phonetik.uni-muenchen.de/~jmh/papers/aps.pdf)    

The core idea of Emu is:

- A speech database is a *collection of utterances* consisting of signal (waveform, formant, f0 etc) and annotation (label) files.

- There is *a query language*: annotations can be extracted from the database and read directly into `R`. E.g. find all `[i]` vowels in the database.

- These queried lists of annotations in `R` can be further queried to get the corresponding signals (*find the formants of the `[i]` vowels extracted at the previous step).

There have been some landmark changes to Emu since the APS-Edinburgh days of the 1980s.

1. In the 1990s (at Linguistics, Macquarie University, Sydney), Emu's query language was updated to handle *hierarchically structured annotations*: see especially [Cassidy & Harrington, (2001)](https://www.phonetik.uni-muenchen.de/~jmh/research/papers/cassidy01.specom.pdf) (The name `Emu` evolved out of  Extended MUlti-dimensional and because we were in Australia).

![](img/emu1990s.png)

The author of the Emu query language is [Steve Cassidy](https://researchers.mq.edu.au/en/persons/steve-cassidy) who can be seen here feeding the Emu ca. 1996. The query language is still in use today and the only one of its kind in existence.

![](img/cassidy.png)


$~$



The point of hierarchical annotations is to be able to query annotations at one tier with respect to another. E.g., the previous query could be extended to:

- find all `[i]` vowels in the first syllable of trisyllabic accented words,  but only if they are preceded by a function word in any `L%` intonational phrase.

2. From 2002-2006, the ASSP signal processing toolkit developed by Michel Scheffers of the IPdS, University of Kiel was integrated into Emu ([Bombien et al,2006 ](https://www.phonetik.uni-muenchen.de/~jmh/papers/emusst06.pdf)). `ASSP` then morphed into `library(wrassp)` ca. 2014.

3. In the last few years, the Emu engine was completely overhauled by Raphael Winkelmann with many excellent new features (See also [Winkelmann et al, 2017](https://www.phonetik.uni-muenchen.de/~jmh/papers/emucsl.pdf)) e.g.:

- Emu is launched and operates entirely within the `R` programming environment. 

- an interactive graphical user interface for analysing and visualising data: the `Emu-webApp`

- extension of the query language to include regular expressions.

- far more rapid access to extracting annotations and their signal files from the database.

# The Emu-SDMS
![](img/emu_overview.png)