(Note: please read all the way through this section before you begin recording.)
Once you have created a new inventory file, the first sentence to be recorded is displayed in the InvTool window, as shown below, and the sentence is spoken aloud for you.
This audio prompt is intended to guide you in how to pronounce the sentences. You can have the prompt be repeated at any time, as often as you want, by clicking the "Prompt" button.
Before you start recording, click on the "Tools" menu and make sure that the "Mic Calibration" mode is not turned on (there should not be a check mark beside it). If it is, click on "Mic Calibration" to turn it off (removing the check mark).
When you are ready to record the sentence, click the "Record" button, pause a half-second, and begin speaking. Note that the "Record" button changes into a red "Stop" button, as shown below. When you are finished speaking, pause another half-second, then click the "Stop" button; it will change back to the black "Record" button.
BEWARE: Many people have a tendency to begin speaking right as they click the "Record" button, and to click the "Stop" button right as they finish speaking. This may result in the beginnings and ends of your recordings being chopped off. To prevent this from happening, it is important to pause briefly before and after speaking, as mentioned above. This pause needs to be only about a half-second long.
The Performance Meters
After you click the "Stop" button, watch the three Performance Meters at the bottom of the InvTool window. The purpose of these three meters is to ensure that your recordings remain as similar to each other as possible throughout the recording process. This uniformity is essential to making a high-quality voice.
If your recording is within the acceptable ranges, the boxes underneath each meter will say "Good!", as in the above picture. Your recording will be automatically saved, and InvTool will proceed to the next sentence.
However, if your recording is not within the acceptable ranges, the boxes underneath the meters will say "Too High!" or "Too Low!"; "Too Loud!" or "Too Soft!"; and/or "Unclear", as appropriate, as shown in the picture below:
When this happens, simply record the sentence again, as many times as necessary, until your recording is within the acceptable ranges.
Setting The Target Pitch: Pay careful attention to your pitch as you record the first sentence in your inventory. The Pitch Performance Meter assumes that the pitch of your first recording is your target pitch, and compares the pitch of all of your subsequent recordings to this one. Therefore, it is important that the pitch of your first recording indeed be as close as possible to your normal speaking pitch. If it is not, you may find yourself having to speak in an abnormally high or low pitch as you continue your recordings.
Adjusting Performance Meter Calibration:
Loudness and Pitch Meters: The calibration of the Loudness Meter was described in Section 2, and the calibration of the Pitch Meter was described in the previous paragraph ("Setting the Target Pitch"). If you continue to have difficulty with these meters, you may try to adjust their calibration settings by hand, as described in Section 9. However, because the purpose of these meters is to keep your pitch and volume as uniform as possible throughout the recordings, it is best if you use the same parameters for these meters throughout the recording of each inventory.
Pronunciation Meter: One of the most important functions of InvTool is to locate each speech sound (phoneme) within the speech you record. It uses speech recognition to do this and, while the speech recognizer is fairly robust, it makes errors. If your speech is somewhat weak, breathy, hoarse, or slurred, the recognizer may make quite a few errors. Unfortunately, the Pronunciation Meter depends entirely upon the speech recognizer to determine pronunciation accuracy. When the recognizer functions correctly--that is, when it correctly locates all the phonemes in your recording--the pronunciation rating is (we believe) a useful measure of pronunciation. However, if the recognizer makes errors in locating the phonemes in your speech, the Pronunciation Meters is likely to indicate that you mispronunced the utterance whether your pronunciation was actually quite good or not. Thus, when you record an utterance, the Pronunciation Meter is either providing a measure of your pronunciation or a measure of the success of the recognizer, but it is not always obvious which! Fortunately, despite this ambiguity, the best solution for either a true mispronunciation or a poorly recognized correct pronunciation is to re-record the utterance. We should also note that if you record an utterance that is quite different than the one InvTool expects, the Pronunciation Meter reading is very unreliable.
We suggest the following guidelines for using the Pronunciation Meter: If you record any sentence perfectly twice, and the Meter rejects it both times, go ahead and "Save" the recording (as described below) and move on. If you continue to have frequent difficulties with the Meter, even after carefully following the guidelines in Section 7, you may adjust this meter's parameters or turn it off entirely, as described in Section 9. If you do this, however, remember that you are increasing the risk of accidentally recording something that does not match the audio prompt and/or including recordings with incorrect phoneme locations that will disrupt the speech synthesis program.
The "Save" Button:
The "Save" button allows you to save a recording that is not within the acceptable performance meter ranges, and move on to the next sentence. You may then come back to the sentence at a future point and re-record it (see the next section).
Please Note: By default, a recording outside the acceptable ranges that is saved this way will be ignored by the program when it creates your synthetic voice (Section 10). In other words, the program acts as if you never recorded it.