Wolfgang Hess 60 years young

1
Wolfgang Hess 60 years young
Speech is beautiful

• Louis C.W. Pols
• Institute of Phonetic Sciences
• University of Amsterdam

Bonn, Sept. 29, 2000
2
IKP, Bonn
IFA, Amsterdam
3
Speech is beautiful

• most natural form of communication
• it is efficient
• highly complex and challenging
• towards multi- and interdisciplinary communities
• natural speech synthesis ? full knowledge
• ASR lasting challenge
• speech is extremely robust to distortions
• speech is eloquent singing speeches are awful
• speech community is nice

4
robustness to degraded speech

• partly reversed speech
• (Saberi Perrott, Nature, 4/99)
• fixed duration segments time reversed or
• shifted in time
• perfect sentence intelligibility up to 50 ms
• (demo every 50 ms reversed original )

5
Wolfgang

• engineer by training
• emphasis on signal processing (Münich)
• pitch-synchronous spectral analysis
• applied for phoneme and word recognition
• and for voice detection and pitch extraction
• speech synthesis (Bonn)

6
History, almost 30 yrs ago

• 7th International Congress on Acoustics 1971,
  Budapest, Hungary
• first international (speech) conference
• Satellite Speech Symposium, Szeged
• Hess, Grundfrequenzsynschrone digitale
  Spektralanalyse von Sprachsignalen mit beliebig
  feiner Auflösung im Frequenzbereich
• also papers in German, and even in Russian
• engineering interest in speech analysis
• forthcoming specialization in sp. recogn. pitch
  extr.

7
Budapest ICA

• many influential people from international speech
  science community, already present there
• topics at that time far away from our present
  interests in almost every respect
• topics and ambitions
• approaches taken
• type and size of data sets
• see some names and topics (nostalgia!)

8
speech processing

• Velichko (Russia) dynamic programming
• Bishnu Atal towards predictive coding
• Sakoe (Japan) dynamic processing for time
  normalization
• Osamu Fujimura
• - dynamic palatography,
• - electromyography (hooked-wire electrodes),
• - computer-controlled dynamic radiography
• (Tokio x-ray microbeam generator)
• Jim Flanagan focal points in sp. comm. research

9
speech synthesis

• Cecile Coker articulatory synthesis
• Paul Mermelstein Bishnu Atal vocal transfer
  functions for speech synthesis
• Johan Liljencrants formant synthesis OVE III
• Helmut Mangold synthesis with a limited set of
  dynamic transitions
• Werner Endress synthesis via intermediate sounds
• Peter Denes word concatenation
• Fujimura, Coker Umeda prosody in synthesis
• Larry Rabiner 2-pole digital filters for
  synthesis
• we were away a year ago

10
speech recognition

• Hans Tillmann (abs.) Bonner DAWID-II-system
• Kasya, Kido, Krause Tarnóczy vowel recogn.
• Velichko 60 words
• Rao 225 VCV utterances, diad matching
• Sakoe 2300 isolated Japanese 10 digits
• Dreyfus-Graf artificial language
• Erman 54 isolated words over telephone
• Neeley 54 words recognition in noise
• Pols 50 Dutch words, stationary phoneme parts
• Renato de Mori zero crossings

11
speech perception, musical acoustics,
psycho-acoustics

• Rao plosive-vowel interaction
• Kozhevnikov AM vowel-like stimuli
• Ludmilla Chistovich vowel discrimination
• Johan Sundberg pitch extraction of folk music
• Max Mathews music synthesis
• Tammo Houtgast lateral inhibition in psychoac.
• Evans Wilson neurophysiological evidence
• Bela Julesz critical bands in vision and
  audition
• Egbert de Boer reverse-correlation method

12
Wolfgangs further carrier

• Dissertation in 1972
• Digitale grundfrequenzsynchrone Analyse von
  Sprachsignale als Teil eines automatischen
  Spracherkennungssystems
• Masterpiece in 1983, 698-pages book
• Pitch determination of speech signals.
  Algorithms and devices, published by Springer
  Verlag.
• Chair in Phonetics in Bonn in 1986
• publications, keynotes, conference organizer

13
ESCA/ISCA and Eurospeech

• ESCA grounded in 1988
• Joseph Mariani first president (1988-1993)
• Louis Pols 2nd president (1993-1997)
• Wolfgang final keynote at E97 in Rhodes
• since Sept. 1997 Roger Moore president
• since death Christian Benoit (April 25, 1998)
  Wolfgang secretary of ESCA
• since Eurospeech99 in Budapest ? ISCA

14
ICA 1971

• all speech analysis based on filters or formants
• LPC was about to be introduced
• all synthesis based on formant synthesis
• diphone concept did not yet exist
• virtually no attention for TTS synthesis-by-rule
• all speech recognition based on word-template
  matching
• probabilistic approach yet unknown
• vocabulary size of the order of 50 words only

15
present-day synthesis

• mainly corpus-based concatenative synthesis with
  non-uniform units (e.g., CHATR, Festival,
  Next-Gen, Laureate, Bonner system)
• large storage, optimal search
• high naturalness and intelligibility
• but.one speaker, one style, one application
• room for further improvement

16
possible improvements

• general or application-specific corpus
• how to reduce storage requirements
• annotation details at various levels
• optimize search algorithms and cost functions
• fewer prototypes, generate certain variants
• preferable units, fall-back mechanism
• new voice, speaking style, emotion, rate
• can voice be personalized (cont.)

17
possible improvements (cont.)

• how much manipulation in concatenation
• combining stored speech and synthetic speech
• better prosody (copy, concept, rules)
• intonation modelling (discrete or continuous
  detailed or sparse signal oriented or
  linguistically meaningful)
• concept for duration modelling
• sentence accent and prominence

18
presently not very popular

• formant synthesis (but see MITalk)
• diphone and demisyllable synthesis (but see many
  operational systems Dutch Fluency, German
  Hadifix, Multi-lingual Lucent TTS)
• use of forms of parameterized speech (as soon as
  more manipulation is required again)
• many voices, speaking styles, emotions, rates
• importanc of system evaluation (Jenolan Caves)

19
future for Wolfgang

• being in the midst of new and challenging
  developments
• to produce
• (in the most efficient way)
• the highest achievable quality of synthetic
  speech
• (given specific dialogue applications)
• is a large responsibility
• but also a lot of fun to do (cont.)

20
future for Wolfgang (cont.)

• Wolfgang and the IKP group enjoy doing this
• for German and other languages
• and like to report about it at international
  forums
• it attracts many good students
• these are excellent conditions for continuing
  this work

? I wish Wolfgang and all his colleague a lot of
success in the years to come!