Title: MultiLing Project
1Proof of Concept
- Multi-Ling Project
- Ray Carrick
- George Bett
- Alan Blunt
- Keith Edwards
- University of Edinburgh
- SE Ref 6-CIS004
2What is the Multi-Ling project?
- 8 years UoE research on text generation from
symbolic object descriptions - M-PIRO (EU) project in museum domain
- English, Italian, Greek languages
- texts for children, adults, adult experts
- sensitive to viewing history
- small-scale database (100 artefacts)
- not commercial grade performance
- Multi-Ling PoC
- scale up to large databases (200,000 artefacts)
- real-time performance for multiple users
- market assessments in three key sectors
3M-PIRO
4PoC Achievements
- Key technical objectives met
- Improved robustness
- more efficient database system (persistent data)
- better algorithms
- Improved scalability
- reduced memory/disk usage per request
- multi-threading support
- improved response time per request
- Improved maintainability
- regression testing environment
- modular interfaces
- open architecture (OpenCCG) for grammar engine
- Market Assessment identified online music domain
5Application architecture Museum domain
User museum visitor
commands prev/next etc.
user settings class of user language viewing
history
metadata museum artefact facts etc.
language data museum domain grammars, vocab
text generation engine
artefact descriptions
6Application architecture Music domain
User content programmer for internet radio
user settings lang/style prefs themes voice
settings
metadata AllMusic Guide track/artist facts etc.
playlist
commentary text stream
audio mixing
language data music domain grammars, vocab
text generation engine
music
spoken introductions, track links
text to speech synthesis
7Market Structure Online Music delivery
- Broadcasters
- internet radio vs simulcast
- e.g. Live365, Shoutcast, AOL Radio
- Narrowcasters
- personalised/ customised streams
- e.g. Pandora, Last.fm (CBS), Slacker, launchcast
- Retailers
- subscription vs unit sales
- e.g. eMusic, Rhapsody, Napster, Amazon, AOL,
iTunes
End User PC, Mac etc. Custom portable
device Mobile phone
End User PC, Mac etc. wi-fi internet radio
End User PC, Mac etc. Custom portable
device Mobile phone iPod/mp3 player
8Value proposition for broadcasters
- The user is the content programmer
- Input a playlist
- Process setting preferences for style, themes,
amount of text, choice of voice etc. - Output
- if text-to-speech is used
- a new playlist with audio introductions and links
between tracks for webcasting or download
(podcast) - DJ4me adds a voice to jockless radio
- could impact traditional broadcasting c.f.
JackFM - if not
- relevant text streams make the webpage stickier
- powerful attractant to advertisers
9Value proposition for narrowcasters
- (A narrowcaster produces a unique stream for each
listener) - The user is the end listener
- Input an individual song (selected by the host
software, not the user) - Process setting broad preferences, rating and
reacting to songs, i.e. user interaction - Output
- if text-to-speech is used
- audio introductions and links between tracks
- personalised commentary on ratings etc, i.e. make
the user interaction come alive! - if not
- relevant text streams, updated user profiles
10Value proposition for retailers
- The user is the end listener
- Input an individual song or a playlist
- Process some user preferences could be offered
- Output (assume text-to-speech is used)
- audio introductions for individual tracks
- links between tracks for podcasts/multiple
downloads - Opportunities for retailers
- differentiate add value with audio intros/links
related to downloads - offset download price with advertising revenue
and target audio adverts at individual users
11Feature focus
feature
12 months
beyond
6 months
- more languages
- more domains
- advertising?
- games?
- non-music
- metadata
- question/answer
- pairs
- intros/links
- commentary
- auto-updates
text generation from music metadata
- domain/market
- specific
- voice
- editing/morph
- controls?
- browser based
- playlist tool
- user controlled
- text length
- text preview
- media player
- user controlled
- text styles, voice
- characteristics
- non-music themes
user configuration via graphical interface
- range of
- languages
- custom voices
- text-to-speech
- voice integrated
speech output
- narrowcast/retail
- (multi-threaded)
- product vs.
- service options
- broadcast
- (single threaded)
system configuration
12Project outline
task category
milestone 1 (mid Nov)
milestone 2 (end Jan)
milestone 3 (end Feb)
- lang. generation
- validated
- user interface
- completed
- -system integrated
- system tested
- speech output
- optimised
- music domain
- lang. generation
- initial user
- interface
Technology
- market report
- analysed
- demonstrator
- features known
- -key partners
- identified
- Marketing strategy
- developed
- key partners
- contacted/NDA
- -demo website
- designed
- demo website
- up running
- -key messages
- -partner feedback
Marketing
- feature roadmap
- funding options
- resource planning
- potential license/
- spinout options
- -angel pitches
- IP negotiations
- resource planning
- infrastructure
- planning
- spinout/licensing
- options identified -staff/resources
- IP license
- Infrastructure
- funding options
Commercial-isation
13Conclusion
- solid, scaleable technology
- large, growing market
- technology enhances online music delivery
- 6 months to commercialisation
- The following slides were produced to support a
general discussion about the broadcast sector of
the online music market
14US Internet Radio Growth Projection
monthly audience (millions)
source www.bridgeratings.com
15UK Internet Radio Growth
24.3 of population have used internet radio
source OFCOM
16US Internet Radio Advertising Trends
millions
source www.bridgeratings.com
17SoundExchange/CRB webcasters
- March CRB royalty rates 2006-2010
- 3x increase in per play new 500/channel fee
- 150-200 of total revenues for typical webcasters
- April Internet Radio Equality Act bill
introduced in House Senate - April SoundExchange negotiates w/ NAB/webcasters
- May SoundExchange offer to small webcasters
- small gross annual revenue
- 2002 Small Webcaster Settlement Act extended to
2010 - Royalty 10 revenue up to 250k, 12 beyond
- August 23 agreement with large webcasters
- 50k cap on per-channel fee
- consensus reporting, anti-stream-ripping focus
- August 29 royalty rates for large webcasters
still to be resolved
18University of Edinburgh