Planning of Recruitment

The last thing to do before going into the collection phase is the preparation of the speaker recruitment.

Again, the recruiting technique you'll use depends on the kind of speech corpus you're producing and on how much funding you have. You will find general hints from our long experiences with different corpus productions in section (p. ). Here is some advice as to why you might need to think about the recruitment way ahead of the start of the collection:

• If you have plenty of funding, you might consider to out-source the problem to an advertising or market research agency. Some agencies keep extensive databases that may be used to find speakers of your desired profile and mail them directly. Expect costs of about EUR 20-40 per successfully recruited speaker (without the incentive). We mention that here already because these agencies usually need some time to prepare (2 months).
• If you do the recruitment yourself, assign one person from your staff to it. This person should then take care during the collection phase that enough speakers of the needed profiles are scheduled for the recording slots. Give that person some time to get familiar with the problem
  (1 month).
• Usually it's a good idea to start the recruitment at least a month ahead of time. Your best asset in the recruitment business are the recruited speakers themselves. Offer them more incentives for each new recruited speaker (snow ball systems) and they will go for it like starved squirrels for the roasted peanuts. But to get a significant mass this takes some time. Therefore it is good to start early and do a lot of pre-scheduling.

Please also refer to the chapter (p. ) for legal advice in the matter of storing speaker information.

It is a good idea to collect data about one's speakers in a database [2], pp. 138:

``Speakers should be thought of as a primary and very valuable resource in speech recordings. It is therefore advisable to build a speaker database which contains for each speaker

• a unique speaker id,
• administration data (name, address, telephone),
• personal information (place and date of birth, languages, education, etc.),
• physiological data (sex, size, weight, etc.),
• speaker history (list of recordings, etc.),
• remarks.

The recruitment of speakers should have two goals: provide a sufficient number of speakers for a given speech data collection, and provide sufficient information about the speakers which can be used to build or extend a speaker database.''