Signal Processing and Speech Communication Laboratory

GRASS - Data Collection

This article is part of GRASS: the Graz corpus of Read And Spontaneous Speech.

Equipment

HM
close talking head sets (AKG HC-577)
FM
large diaphragm microphone (AKG C-414) + pop screen
LG
laryngograph → F0 reference
V
video camera
LAWO
mixing desk

Sound Quality

  • 16 bit, 48 kHz sampling rate
  • average SNR over all speaking styles of 49.6dB
  • average SNR in conversational speech of 46.4dB

Recording Procedure

(1) Conversational Speech

1.1 Speakers (two friends or two family members) enter the reception area: contracts and instructions.

1.2 Speakers enter recording area. Instruction: talk about whatever they want; they might use cards to start the conversation.

1.3 Experimenter leaves the room, returns in one hour.

Break.

(2) Elicited Commands

2.1 Speaker 1 enters recording area.

2.2 Pictures on screen presented with SpeechRecorder elicit e.g., "Bitte die Tür schliessen!"

(3) Read Speech

3.1 Speaker 1 reads sentences and digits from screen.

Break.

2.1 -- 3.1 for Speaker 2.


Further Reading on GRASS