The Centre for Speech Technology Research, The university of Edinburgh

Recording and labelling procedure

Recording procedure

As shown in the design of the sentence materials (see DESIGN page), there are five keyword series for each test syllable, each containing three sentences. The three utterance-medial series have one sentence in common, as do the two utterance-edge series, making a total of twelve sentences for each keyword, each to be read in two accent conditions, as illustrated in the example sentences (see EXAMPLES page).

There are eight right-headed keywords and eight left-headed keywords, making 384 experimental sentences in total, listed according to keyword series in the full set of sentences (see SENTENCES page). Each experimental subject recorded two repetitions of each sentence, making a total of 768 sentences recorded by each subject and 4608 recorded sentences in total.

Recordings were made direct to disk in ESPS format at a sample rate of 16 kHz, using a Sennheiser MKH815T shotgun RF condenser microphone with AB powering and a 24dB front-to-back ratio in a soundproofed recording studio. Further details of the recording procedure are available on the following pages.

Labelling procedure

Following the organisation of recorded sentences into sound files and the alignment of identifying labels with each utterance, a three-stage labelling procedure took place. Firstly, utterance production labels - for example, identifying incorrect placement of emphasis - were applied as appropriate. Secondly, the start and end points of the onset, nucleus and coda of each test syllable were identified by inspection of waveforms and spectrograms. Finally, the start and end points of the additional syllables within polysyllabic keywords were identified for a large subset of the utterances. The labelling conventions used in each of these stages is described in detail on the following pages.

Up to EUSTACE HOME page Forward to RESULTS page