Analysis of the New Zealand Emglish and Mori OnLine Translator

1
(No Transcript)
2
Development of a Mäori Database for Speech
Perception and Generation

• Mark Laws
• Knowledge Engineering Laboratory
• Department of Information Science
• University of Otago
• Dunedin
• New Zealand

3
Content

• An introduction to the Mäori language
• Mäori language history and development
• Speech data collection and analysis - 2 views
• Focus on the Diphone speech unit
• The Mäori diphone inventory - tools
• Mäori Speech Synthesis - examples
• Prosody - speech parameters
• Summary and additional information

4
Introduction

• The Mäori language is part of the
  Proto-Eastern-Polynesian group
• Mäori is only spoken in New Zealand (Aotearoa)
• Mäori speech data collection and analysis is part
  of the Otago Speech Corpus Project
• Mäori speech and language processing is a new
  research development theme

5
The Mäori Language

• The language has undergone many changes
• Since early contact the language was influenced
  by English
• More recently, the language was in crisis of
  survival
• Now teaching the language at all educational
  levels has improved its long term survival
  chances
• Mäori is now an official language of New Zealand
• Mäori is used in all aspects of social and
  cultural exchange
• The language is different from English
• The phonemic systems are different (vowels are
  constant)
• There are less phonemes (24M, 45 NZE)
• The phonotactics are simple (e.g. CV - C is
  optional)

6
(No Transcript)
7
(No Transcript)
8
Speech Analysis

• Computational linguistic and Information
  processing techniques applied
• Unit Segmentation
• Phonetic Labeling
• Acoustic characteristics
• Formant analysis
• Prosodic modeling
• Feature extraction
• Database management
• Phone and Diphone analysis

9
The Diphone

• All the phonemic transcriptions taken between
  all the possible phonemes used in a language
  (e.g. over 1400 for English, less than 300 for
  Mäori)
• Contains the phonemic transitional boundary
  between two phonemes, it holds the rich
  co-articulation information that is essential for
  natural speech
• Is segmented from speech by identifying the
  boundaries between phonemes within word
  utterances
• Concatenative speech synthesis uses the diphone
  as the fundamental unit of selection

10
Mäori Diphones

• Inventory taken from 229 selected words that
  contain all possible diphone combinations
• Native speaker chosen, speech recorded at high
  quality, constant pitch is maintained
• Diphones are hand segmented then labeled using
  the SAMPA notation (e.g. A-e, N-A, f-aU)
• Post-processing to compile the diphones into a
  pre-specified database called mb1

11
(No Transcript)
12
(No Transcript)
13
Mäori Synthesiser

• The diphone database was tested with tools
  designed as part of the MBROLA Project. The
  projects main aim is to provide a multi-lingual
  speech synthesis system under the open source
  protocol
• MBROLIGN and MBROLI are tools to test and
  generate files that are required to produce high
  quality speech synthesis
• The Festival text-to-speech system is compatible
  with the diphone database, it is used to complete
  the Mäori Speech Synthesiser as a fully
  functional system on-line

14
(No Transcript)
15
(No Transcript)
16
Speech Examples
Mäori speaker - recorded word kete Diphone
h-i taken from the word hia Mäori speech -
joined diphones kia ora Mäori synthesis -
diphones (1) kia ora Mäori synthesis -
diphones (2) kia ora English synthesis -
diphones kia ora Mäori speaker -
rulebased kaka
17
Prosody

• Language models based on speech parameters
• Duration
• Pitch
• Stress
• Voicing
• Intonation

18
Summary

• Step for the Mäori language to become global
• Speech database management important basis
• Speech data signal analysis
• Research using the diphone in speech recognition
• Future speech apps implemented using Java

19
Additional

• Datasets http//kel.otago.ac.nz/maaka/dataset
  s/
• This paper http//kel.otago.ac.nz/maaka/datasets/s
  ynthesiser.pdf
• On-line paper http//kel.otago.ac.nz/maaka/dataset
  s/synthesiser.html
• This presentation http//kel.otago.ac.nz/maaka/dat
  asets/synthesiser.ppt

20
Acknowledgements

• Information Science and Mäori Studies Colleagues
• New Zealand Foundation for Research Science and
  Technology - CBIIS Grants and Fellowships
• JCIS 2000 Committee

21
(No Transcript)