Next: Recording Techniques
Up: Preparation of collection
Previous: Preparation of collection
Contents
Instructions and Prompting
All speakers participating in a speech corpus recording need some kind of
instruction before the recording starts. The instructions may range from a
very reduced instruction set in psychologically inspired experiments or Wizard-of-Oz
recordings to very detailed and strict instructions for an
unsupervised collection over the telephone network. Although possible,
we strongly discourage you from relying only on a verbal instruction given by
the supervisor or experimenter before the recordings. To improve the
consistency of the recordings always use a written instruction, or
use pre-recorded instructions.
Make the instructions as simple and as unambiguous as possible. Don't
overload them with background information but give a brief outline
of what the collected data is going to be used for. Outline the contents,
the speech style, the recording technique and the estimated length of
the recording.
In many speech collections you will prompt the speaker to produce specific
utterances. This can be done acoustically (resulting in mimicked speech
styles) or on paper or a monitor (resulting in read speech). Except for special
purposes where a certain prosody, loudness or emotional speech style
should be elicited we strongly recommend using written prompt material.
You may also use direct questions -- again acoustical or written -- to
elicit certain utterances (answering speech), ask for descriptions
of an image or a video movie (descriptive speech) or give directions for a
monologue or dialogue (non-prompted speech).
In any case formulate your prompts so that they are un-ambiguous with
regard to phonemic form and stress. If you are prompting with questions,
restrict the number of possible answers:
``What is your favorite dish? (name one)''
Technically the prompting can be done in several different ways:
- Let the speaker read the prompts or questions from
paper5.1.
- Display the
orthographic form of the utterance or the question on a screen.
- Display of images or video movies.
- Playback of pre-recorded acoustical prompts or questions.
- Direct questions from the experimenter (interview).
- Combinations of these.
Depending on the recording setup you may use an automated process that
controls the recordings within one recording session. For telephone
recordings (see below) using a telephone server this is a must, but also
in most other types of recording setups we recommend using some script
language to control the recording times, the signal file names, the
questionnaires for the speaker etc.
Here are some practical hints for the setup of your recording procedure:
- If you are using a paper version and an automated process
simultaneously (like in
a telephone recording), be sure that they are consistent. Nothing
confuses a speaker more than inconsistent instructions.
- Use some dummy recordings at the beginning of the session and
possibly also at the end. You may announce them explicitly as training
prompts.
- Carefully design the order of prompts. Avoid sequences of similar
utterances, e.g. many sequences of numbers. This may result in a
droning speech style.
- Use a beep or a visual marker (for instance a `red light') to
indicate when a recording begins or ends.
- Give feedback to the speaker. For instance:
``You have already finished more than half of the recordings!''
Test your procedure, including instructions set, prompt material
and automated recording program on naive speakers before starting the
collection.
Next: Recording Techniques
Up: Preparation of collection
Previous: Preparation of collection
Contents
BITS Projekt-Account
2004-06-01