Use case: long video interviews with interlocutor speech, noise, long silence intervals etc.

What we have: Long sound track (longFiles/UniMuenster.wav) and the orthographic transcript of target speaker in form of a 'chunk segmentation' (longFiles/UniMuenster.TextGrid), in which the important parts or the target speakers are orthographically transcribed (and segmented into so called 'chunks') in a single annotation tier (the tier name is 'Transcription 1' in this example).

What we want: Complete segmentation of target speaker by words and phones

Solution: Web Interfaces Chunk Preparation + G2P + WebMAUS General

* go to http://clarin.phonetik.uni-muenchen.de/BASWebServices

* go to service Chunk Preparation

* upload 'longFiles/UniMuenster.TextGrid' (drag&drop the file to the upload area and click 'Upload')

* execute the Chunk Preparationi service with the following options (all other options default):

    Language = English (GB)
    Input tier name = Transcription 1
    Sampling rate = 16000

  The log area should turn green; if not green, check the ERROR/WARNING messages (click on a message to expand).
  What this service does: it transforms a praat TextGrid annotation into a BAS Partitur Format file which can be
  processed by the BAS Services.

* download the resulting file 'UniMuenster.par' to the local dir longFiles/

* goto service G2P

* upload 'longFiles/UniMuenster.par' (drag&drop the file to the upload area and click 'Upload')

* execute the G2P service with the following options (all other options default):

    Language = English (GB)
    Input format = bpf
    Tool embedding = maus

  What this service does: it reads the text chunks from the annotation tier, tokenize and normalize the text, and then
  calculate the most likely pronunciation transcript for each word (encoded in SAMPA).

* download the resulting file 'UniMuenster.par' to the local dir longFiles/ (overwriting the existing file)

* goto service WebMAUS General

* upload 'longFiles/UniMuenster.wav' and 'longFiles/UniMuenster.par' (drag&drop both files to the upload area and click 'Upload')

* execute WebMAUS with the following options (all other options default):

    Language = English (GB)
    KAN tier in TextGrid = true
    ORT tier in TextGrid = true
    Chunk segmentation = true
    Output symbols = ipa

  What this service does: based on the phonemic, most likely transcript (from G2P) it calculates a statistical predictor model 
  about possible phonetic realisations of the recording, then combines this model with acoustical models of the individual 
  phones of this language and derives the most likely phonetic realisation in form of a labelling & segmentation.

* download the resulting *.TextGrid and open it with praat, or
  check result directly in your browser using the EMU webApp (click on the result link, and then click on the blue box EMU webApp symbol in the preview box)

[Note: you can perform all three processing steps described above in one step by using the service 'Pipeline w/out ASR'. Upload 
the files 'longFiles/UniMuenster.wav' and 'longFiles/UniMuenster.TextGrid' and start the service with options:
  Pipeline name = CHUNKPREP_G2P_MAUS
  Language = English (GB)
  (Click on 'Expert options')
  Output symbols = ipa
  Input tier name = Transcription 1
]

