Corpus of European Portuguese Stops

Introduction

A corpus of 54 European Portuguese real words containing /p, b, t, d, k, g/ was designed to study the temporal and spectral characteristics of stop consonants.

Corpus 1 and 2

The corpus contained an equal number (eighteen) of words with stops in: initial position, followed by the vowels /a/, /i/ and /u/; medial position, preceded by the vowels /a/, /i/ and /u/ and followed by the vowel /ɐ/; final position, preceded by the vowels /ɔ/ and /a/. The words were produced without any context (Corpus 1) and within the frame sentence "Diga,... por favor." (Corpus 2).

Recording Method

The subjects were six native speakers of European Portuguese, three men LJ, HR and PA (aged: 25 to 34) and three women ML, IM and SC (aged: 24 to 42), all without any speech, language or hearing problems. Subjects LJ, HR, ML and IM are from the city of Aveiro (at the centre of Portugal) and Subjects PA and SC are from Porto (in the north of Portugal).

The corpus was recorded using a Philips SBC ME 400 unidirectional condenser microphone located 20 cm in front of the subject's mouth. A laryngograph signal (Lx) was also collected using a laryngograph processor (model EG-PC3 produced by Tiger DRS, Inc., USA). The acoustic and Lx signals were pre-amplified (Rane MS 1-b) and recorded with a Sony PCM-R300 DAT recorder, each with 16 bits and a sampling frequency of 48 kHz.

Segmentation and Annotation Method

The corpus was manually segmented with Adobe Audition. The words were then analyzed using Speech Filling System (SFS) and the following acoustic events were annotated:

1. The beginning of preceding vowel (IV1) was defined as the instant in time at which the second formant intensity becomes characteristic for a vowel.
2. The end of preceding vowel and beginning of closure (IO) was marked where the second formant was no longer visible.
3. The beginning of prevoicing (IPV) was defined as the instant in time at which evidence of vocal fold vibration could be observed both in the acoustic and Lx signals.
4. The voice offset (FV) was marked at the point where the vocal folds ceased to vibrate, as can be seen in the Lx signal in Figure 1 (the periodic signal ends).
5. The end of prevoicing (FPV) was defined as the instant in time where the burst started.
6. The end of closure and the beginning of the release (IR) was defined by a sudden peak in the waveform and as a vertical bar in the spectrogram. When there were multiple bursts, the one with the highest intensity was chosen.
7. The end of release and beginning of following vowel (FR) was marked where the second formant amplitude reached typical values for a vowel.
8. The end of the following vowel (FV2) was set where the second formant was no longer visible.

• When the duration of voicing was absent, the voiceless stops (0).
• When voicing was present during < 1/3 of the closure interval, the stop was classified as devoiced (1).
• When voicing lasted from 1/3 to 1/2 of the closure duration, the stop was classified as partially devoiced (2).
• When the duration of voicing was > 1/2 of the closure duration, the stop was classified as voiced (3).

Corpora Documentation and Distribution

Complete listing of all recorded material can be found in the M.Sc. Thesis by Lousada (2006).

The corpora can be stored in one CD-ROM that contains the following data:

• The .wav files contain the acoustic speech signal in the left channel and the laryngograph signal on the right channel, recorded at 16 bits, with a sampling frequency of 48 kHz.
• The files with the extension .seg are annotation files. In Corpus 2 there are 10 time values (in seconds) referring to:
  1. The beginning of preceding vowel
  2. The end of preceding vowel and beginning of closure
  3. The beginning of prevoicing (IPV)
  4. The voice offset (FV)
  5. The end of prevoicing (FPV)
  6. The end of closure and the beginning of the release (IR)
  7. The end of release and beginning of following vowel (FR)
  8. The end of the following vowel (FV2)
  9. Position in word
  10. Type of voicing
• The .ipa files contain a SAMPA phonetic transcription of the speech segments contained in the .wav with the same name.

Back to Homepage