Find topic
WS06 topics
Members' area
Tools
Help!
-- KarenLivescu - 15 Dec 2005
|
Final presentation
- Here is a repository of all of my slides from team meetings/update talks--feel free to swipe as you see fit. -Karen
Draft outline
- Introductory material (15min)
- Karen:
- Motivations: Data savings, discriminative modeling, coarticulation modeling, previous results
- Outline: Main models we worked on, preview of the conclusions
- Preliminaries: Feature sets, manual transcriptions, tasks, DBNs, GMTK
- Simon:
- SVitchboard data & baselines (pre-workshop & workshop), gmtkTie
- Hybrid models (15min)
- Simon: Introduction, MLP training, models & results
- Tandem models (15min)
- Ozgur: Introduction, basic models
- Arthur: Factored models
- Multistream feature-based pronunciation models (20min)
- Karen: Introduction
- Chris: Monofeat models & results, tying and initialization experiments, CHMM structure?
- Nash: Cross-word asynchrony modeling
- Lisa: POS-dependent asynchrony modeling
- Bronwyn: Substitution modeling
- Karen: Summary & conclusions
- Audio-visual speech recognition (15min)
- Mark, Partha: CUAVE, models & results (AVICAR?)
- Analysis (15min)
- Nash: Manual transcription & MLP performance analysis
- Lisa: Recognizer error analysis
- Ari: FAFA, comparison of alignments made with different models, wavesurfer tool
- Structure learning (10min)
- Student proposals (30 min)
- Arthur, Partha, Chris, others?
- Summary & conclusions (10min)
- Karen:
- Summary of results & take-home messages:
- Tandem AF models: Beat phone-based models, at least monophone--try them at home!
- Hybrid AF models: Close to phone-based; maybe best for cross-lingual or small-data scenarios?
- Multistream AF-based pron models: Close to phone-based; more work to be done
- AVSR (tentative, based on dev set):
- Multistream helps
- Asynchrony helps
- Articulatory feature is no better or worse than phone-based, but...
- Rover of articulatory feature and phone-based models is best of all
- Other contributions:
- gmtkTie
- Manual transcriptions
- Wavesurfer analysis tool
- New SVB baselines (monophone & triphone)
-- KarenLivescu - 12 Aug 2006
* closing_day_Aug17_06_simon.ppt: Simon's section v2
|
