-- SteveRenals - 29 Jan 2009

• woz_results.ods: This spreadsheet contains the following results:
  • Details of the WOZ corpus (i.e., details of number of utterances and word tokens in young,old,social and Factual categories )
  • Perplexities of the language model built using one category on the test set of another category (viz., old vs young and social vs Factual)
  • Word Error Rates of the test sets in each category using complete Language model, language model built with text from same category and language model built with text from the other category (viz young test set vs old LM and old test set vs Young LM and social test set vs factual LM and factual test set vs social lm). W.E.R's with the following language models are reported a) icsinisthsisl models b) icsinisthsisl models MAP adapted with AMI UK speakers data. and c) Oyez acoustic models.
  • ASR results for individual speaker (of young and old) against the following acoustic models a) icsinisthsisl+AMI(uk) b) icsinisthsisl+AMI(UK) MAP adapted with other WOZ speakers from the same category c) icsinisthsisl+AMI(UK) MAP adapted with WOZ speakers from the other category.
  • ASR results with icsinishsisl + AMI(UK) and AMI language models (50 k word dictionary)

• woz_young_individualAsrResults.csv:

• woz_old_individualAsrResults.csv:

Tasks from the meeting on 29/1/2009

Ravi:

• • set up Wiki
  • write up two experiments
  • edit other literature sections
  • run Oyez voice analysis on MATCH speakers and send files to Maria
  • send .csv file with detailed test results to Maria

Maria:

• • run some more stats on detailed test results
  • mine voice analysis for Interspeech paper
  • walk Ravi through setting up Endnote

14/02/2009 Experimental results with different Language models

• woz_lm_results.ods: Experimental Results with various language models