Transcript Works by Masataka Goto

Works by Masataka Goto

Dr. Masataka Goto (* The photo is taken from Goto’s Home page)

The National Institute of Advanced Industrial Science and
Technology (AIST)
Home page: http://staff.aist.go.jp/m.goto/
Presented by Beinan Li, Music Tech @ McGill, 2005-2-10


Content
Goto’s personal info
 MIR / Music understanding





Speech Interface



Speech Completion
Speech Spotter
Interactive music system




Real-time Beat Tracking System for Musical Acoustic Signals
Real-time F0 Estimation of Melody and Bass Lines in Musical Audio Signals
SmartMusicKIOSK: Music Listening Station with Chorus-Search Function
A Distributed Cooperative System to Play MIDI Instruments
Interactive Performance of a Music-controlled CG Dancer
VirJa Session (A Virtual Jazz Session System)
Music database
Masataka Goto
A researcher working at the National Institute of
Advanced Industrial Science and Technology (AIST), a
newborn Japanese public research organization (15 former)
 A researcher of Precursory Research for Embryonic
Science and Technology (PRESTO) ("Information and
Human Activity" research area), Japan Science and
Technology Corporation (JST)
 Doctor degree from Waseda University, 1998.
 Research interests.

Real-time Beat Tracking System (2001)
(Next)
 Can recognize a hierarchical beat structure
(quarter-note, half-note, and measure levels ) in
real-world audio signals sampled from popularmusic compact discs.
 With or without drums
 Time-signature 4/4 ; tempo is roughly constant
 Using selected musical knowledge (heuristics)
 Succeeded in 43 out of 45 songs
Real-time Beat Tracking System

Main issues of beat tracking from acoustic signal:
detecting beat-tracking cues in audio signals
 interpreting the cues to infer the beat structure
 dealing with the ambiguity of interpretation


Cues:
Onset times of different frequency ranges
 Chord-change possibilities based on provisional time strips
 Drum patterns for Bass/Snare drums

Quantitative rhythmic difficulty: Power transition
 Multi-agent based hypothesis evaluation

Real-time Beat Tracking System

Chord-change possibility, from dominant frequency by histogram
peak within a period of time. (Picture taken from Goto, 2001)
Selectively Used Musical Knowledge



Onset time:
 (a-1) “A frequent inter-onset interval is likely to be the inter-beat interval.”
 (a-2) “Onset times tend to coincide with beat times (i.e.,
sounds are likely to occur on beats).”
Chord change:
 (b-1) “Chords are more likely to change on beat times than
on other positions.”
 (b-2) “….on half-note times than on other positions of beat
times.”
 (b-3) “….at the beginnings of measures than at other positions of half-note
times.”
Drum pattern: (re-evaluate hypothesis)
 (c-1) “The beginning of the input drum pattern indicates a half-note time.”
 (c-2) “The input drum pattern has the appropriate inter-beat interval.”
Real-time F0 Estimation of Melody and Bass Lines
(2004) (Next)
Music Scene Description based on subsymbolic
representation
 Find a predominant harmonic structure instead of a single
fundamental frequency (within a restricted range).
 Melody lines: by a voice or a single-tone mid-range
instrument; Bass lines: by a bass guitar or contrabass
 The average detection rate: 88.4% for the melody line and
79.9% for the bass

Real-time F0 Estimation of Melody and Bass Lines

Main problem:
Which F0 (in polyphonic) -> melody / bass ?
 Unknown number of sound sources.
 Select from several candidates.


Assumptions:
Melody / bass have a harmonic structure, regardless F0
 Melody / bass have a frequency range for most predominant
harmonic structure (“MPHS”)
 Melody / bass line have temporally continuous trajectories (F0),
during a musical note.

Real-time F0 Estimation of Melody and Bass Lines

Method:

Limit the frequency range:
melody : middle- and high-frequency regions
 Bass: low frequency
 whether the F0 is within the limited range or not.


Find the MPHS and its F0


View the observed frequency components as a weighted mixture of all
possible harmonic-structure tone models without assuming the number of
sound sources
Deal with ambiguity

Considers candidates’ temporal continuity and selects the most
dominant and stable trajectory of the F0
Music Listening Station with Chorus-Search
Function (2004)
Music-playback interface for trial listening and general
music selecting / sampling.
 Function for jumping to the chorus section
 Visualizing song structure.


(Picture taken from Goto’s home page)
Speech Completion (2002)
Helps the user recall uncertain phrases and saves labor
when the input phrase is long.
 Based on the phenomenon:


Human hesitates by lengthening a vowel (a filled pause is
uttered): e.g. “Er…”
Displays completion candidates acoustically resemble the
uttered fragment for user to choose.
 Filled pause: small fundamental frequency (voice pitch)

transitions and small spectral envelope deformations.
Vocabulary tree, HMM-based speech recognizer .
 English with Japanese accent? (vowel -> consonant)

Speech Spotter (2004)
Allow user to enter voice commands into a speech
recognizer in the midst of natural human-human
conversation.
 Filled-pause / High-pitch detection based (voice cue).
 On-demand information system for assisting humanhuman conversation (e.g. weather inquiry during talk)
 Music-playback system for enriching telephone
conversation (i.e. BGM judebox)

A Distributed Cooperative System to Play
MIDI Instruments (2002)
 Remote
Music Control Protocol (RMCP)
 Extension
of MIDI. Network symbolized multimedia
information transmission.
 UDP / IP, client-server communication
 Ethernet / Internet
 Information sharing by broadcast and time scheduling
using time stamps
Interactive Performance of a Musiccontrolled CG Dancer (1997)
 CG
character to enhance musician communication
in real jam session via visual attention.
 A successful CG dance depends on interactions
between each musician and CG character.

E.g. If the guitarist plays, CGC does not move unless the
drummer determines the motion timing
VirJa Session (A Virtual Jazz Session
System) (1999)
 Enable
distributed computer players to listen to
other computer players' performances as well as
human's performance and to interact with each
other.
 On top of the last two techniques.
RWC Music Database (2002- )
(back)

RWC (Real World Computing) Music Database
Copyright-cleared
 Common foundation for research. Benchmark.
 Built by the RWC Music Database Sub-Working Group (Goto as
the chair) of the Real World Computing Partnership (RWCP) of
Japan.
 World's first large-scale music database specifically for research
purposes.
 Six original collections, 315 pieces with:

original audio signals
 MIDI files,
 text files of lyrics

individual sounds at half-tone intervals
variations of playing styles, dynamics, etc.
References
 Goto’s
home page:
http://staff.aist.go.jp/m.goto/


Masataka Goto: A Real-time Music-scene-description System:
Predominant-F0 Estimation for Detecting Melody and Bass Lines in
Real-world Audio Signals, Speech Communication (ISCA Journal),
Vol.43, No.4, pp.311-329, 2004.
Masataka Goto: An Audio-based Real-time Beat Tracking System for
Music With or Without Drum-sounds, Journal of New Music
Research, Vol.30, No.2, pp.159-171, June 2001.