HELP

Introduction

The Online Voice Recorder program uses features that are available only in the most recent and advanced web browsers. As of this writing, you must be using the most recent version of Chrome. Neither Safari nor Internet Explorer support this application, and Firefox has not worked well in our tests.

In addition to a compatible browser, you will need a microphone of acceptable quality. Note that unless you are an audio engineer with professional equipment and a studio environment, you must use a head-mounted USB microphone headset to achieve acceptable results. We list several acceptable USB microphone models in our FAQ.

Quick Start for First-time Users

1. Make sure your microphone is connected before logging into the ModelTalker web recorder. You may need to close and re-open Chrome to be sure.
2. If you haven't already done so, login with your ModelTalker username and password, then access the web recorder via the "Recording" > "Online Recorder" menu option.
   • As soon as the recorder page loads, it will try to access your microphone. The first time you visit the recorder page, Chrome will ask you to verify that you will allow the page to use your audio. Of course, you must agree to this! Thereafter, Chrome typically does not ask again unless you have reset it.
3. The Settings dialog box will automatically open.
4. If an inventory is not pre-set, choose an inventory from the dropdown menu (all first-time users must select "Test").
5. Select your microphone or correct input device from the Microphone dropdown.
6. Click Measure.
   • The silence measurement will count down for 5 seconds. You must remain silent while this countdown occurs -- if there is an unexpected noise during this 5 seconds, click Measure again to redo the measurement.
7. Click Listen to hear the silence and make sure there are no sounds recorded.
8. Click Start and begin recording sentences as prompted.

To record a sentence after you hear the prompt, click the REC button, wait for it to change to a STOP button (two vertical bars), and speak. The app will automatically stop recording when you stop speaking and advance to the next sentence; you should not need to click the STOP button.

If you find that the recorder is advancing to the next sentence before you have finished the current sentence, it may be due to an inaccurate silence measure, or pauses in your speech that are longer than the recorder expects. We recommend first redoing your silence measure (reopen the Settings) to be sure it accurately reflects what a silent background is like in your environment. If that does not help, reach out to us for assistance.

After you have recorded four or five sentences, use the Recordings table at the bottom of the screen to review your recordings (click the speaker icon next to each sentence). Do you hear any background noises like fans or heaters or air conditioning? Do the recordings sound "hollow" with echoes from the room? Do you hear any electrical hum in the recording? These are all problems that you will need to correct before you will be able to record speech for voice banking.

If your recordings sound OK to you, finish all 10 sentences of the test set. A red banner will appear above the meters with a "Finalize" button. Click that button to send your recordings to us. We will then check your recordings to make sure your audio system sounds OK, provide feedback to help you achieve a good quality voice, and send you instructions on further recording.

The Details

Display Elements

See this link for an illustrated overview of the online recorder interface, which is described further below.

Menu Items and Progress Bar

In the upper right portion of the screen, below the browser navigation bar, are links to Settings, Voices, Recording Tips, Help Center, and a User menu (showing your username). Click Settings to reopen the Settings dialog. Click Voices to listen to samples of your synthetic voice (see below under "Listening to sample synthetic voices..."). Click Recording Tips to display this web page in a new window or tab. Click Help Center to reach our FAQs and other help documents. Click your username to log out.

Below the menu and above the meters is a box containing theinventory name and a progress bar (visible when recording) that shows your progress toward completing the inventory. A red "Build My Voice" button next to the progress bar (not visible in the Test inventory) notifies us that you want to request your final voice. Clicking this will PREVENT YOU FROM FURTHER RECORDING, so don't click it until you are done with all recording! See "Requesting your final voice..." below.

Meters

Below the progress bar are three meters that will help you monitor your recording to ensure the best quality possible with your hardware. The Peak Amplitude meter measures your loudness, and is the most important meter. The Speech Rate meter indicates how fast your speech is relative to the prompt. If the meter goes to +2 you are speaking twice as fast as the prompt; if it goes to -2 you are speaking half as fast as the prompt. The Pronunciation meter indicates whether the software thinks you have correctly produced the sentence.

See below under "Recording" for more advice about the recording meters.

Controls

Immediately below the meters, the sentences you must record are displayed. Each sentence will also be played in your headphones. After the prompt has played a circular arrow will appear at the end of the sentence text. You can replay the prompt by clicking this arrow.

Below the prompt display area are two buttons:

These buttons are used respectively to (a) start recording the current sentence, or (b) advance to the end of your recordings and show the next unrecorded sentence. Note that there is a specific order to the sentences and the recorder will always follow that order for new recordings. Most of the time, the button to advance to the end of the recordings will be greyed out because you are already at the end of the recorded material and ready to do the first sentence you have not recorded. It will activate if you re-record a sentence in the Recordings Table.

Recordings Table

Below the recording controls is a table that shows all the sentences that you have recorded so far. You can click the Speaker icon next to a sentence to hear your recording, and click on the sentence itself to re-record it. The table is divided into pages, and you can set how many sentences appear on each page. It has multiple other features to allow the user to review their recordings and re-record selected sentences; additional information is provided on our overview page.

How-To's

Getting Started

Before you can record any speech, you must use the Settings dialog to select an inventory (sentence list) and measure the silence (or ambient sound) level in your recording environment. Everyone must start with a short screening inventory called 'Test'. Once you have finished that inventory and we have had a chance to review your recordings, we will provide some feedback and either 1) give you access to the voice banking inventories, or 2) ask you to redo the test inventory. You may be asked to redo the test inventory several times until we are sure that everything is working OK.

The silence measurement is very important. It lets the software know how much noise is in the background (as picked up by your microphone), and that in turn lets the software determine when you have started and stopped speaking as you record each sentence. If the background noise level is too high, you will not be able to do the recording. The silence measurement will be shown as a negative number. Up to a point, the more negative this number, the better, but extremely large negative numbers (-90 to -128) usually indicate a problem with your microphone. The range of values you see depends heavily on the particular microphone and computer you are using, but rough estimates can be given. In a very quiet environment, the silence level should be around -75 dB or better. Values less than about -65 dB are good in most home environments. If you see a value that is higher than -60 dB, you should not try to record.

The voice recorder automatically plays each sentence you need to record and displays the sentence text below the meters. To record the sentence, click the REC button, wait for the button to change to a STOP button (two vertical bars), then begin speaking. When you stop speaking, the voice recorder will automatically stop, save your recording to our server, set the meters to give you feedback, and play the next sentence.

You should check the meters after every recording. Ideally, all meters should end up set to somewhere in their green or yellow range on each sentence, but if the Speech Rate or Pronunciation meters do not, that does not necessarily mean there is a problem with your recording.

• The Peak Amplitude meter is the most important of these. The peak level of each recording must fall within the green range on the meter for each sentence. If your recording is too loud, the program will stop your recording with a warning box and force you to re-record it, but you will need to check that it is not too soft.
• The Speech Rate meter shows your speaking rate relative to the prompt. It is best if your speaking rate remains consistent from one sentence to another. It is okay for your speech to be consistently a little slower than the prompts, but not much slower. If your speaking ability is such that you need to pause between words in a sentence, let us know and we can suggest some advanced settings to help. We advise users not to speak more rapidly than the prompts.
• The Pronunciation meter is the most variable of the gauges. If the pronunciation meter shows a setting that is deeply into the red region of the dial, you should consider trying to back up and redo the sentence, but do not bother to redo a sentence more than twice. It is not at all unusual to find that the meter does not like the way you say some sentences even when your pronunciation is exactly correct (and particularly if your dialect is not American English). So, if two tries does not fix the problem, and you are confident you spoke the words as prompted (the right words in the right order, with no pauses or stumbles), then just move on.

Adding custom recordings

To add custom recordings--e.g. personal names and messages--you must first enter the Custom inventory via the Settings menu. Once you are in the Custom inventory, there will be a link at the top of the page (to the left of the other menu items) to "Add Custom Sentences". Clicking this link will take you a form that will guide you through adding names and phrases that will be added to the end of your custom inventory. You can record the Custom inventory any time after completing the Test inventory, and you can switch between the Standard and Custom inventories at your leisure. You can also use the custom form multiple times.

As you are recording, our system will construct sample versions of your synthetic voice at regular intervals. The first one will be built shortly after you record 25 sentences, the second at 50, and so forth (with progressively longer intervals between new samples). The number beside the "Listen" menu item indicates how many different versions are available to listen to, and will briefly flash when a new version is available. To listen to one of them, click the Listen menu, select a version from the dropdown menu, type any sentence into the text box, and click the button. Note that these samples are built using default parameters; your final voice may sound substantially better than these samples.

Once you have completely finished recording (or decided you can't record anymore), YOU NEED TO LET US KNOW YOU'RE DONE by clicking the red "Build My Voice" button next to the progress bar. This will take you to a confirmation page that will show you the number or recordings you've made in all your inventories, and then ask you to confirm that you're done. Once you've confirmed, we will begin building your final synthetic voice. CLICKING THIS BUTTON WILL PREVENT YOU FROM DOING ANY FURTHER RECORDING, in any inventory, including custom sentences, now or in the future, so you need to be sure you're completely done before clicking it!

Trouble Shooting

Whenever you use the Settings dialog, be sure to check that it is correctly showing your microphone in the Microphone drop down menu. If you don't see your microphone in the dropdown, there may be several reasons:

• No microphones are listed, it just has "Input 1", "Input 2", etc. This is because your audio has never been initialized for our recorder. Select any input and click Measure. After the measurement is finished, close the settings dialog, log out and back in again. The list should then be populated with valid device names. If not, please let us know via email.
• If you connected your microphone after you started the recorder, it might not have been detected. To fix this, leave your mic connected, log out of the recorder and back in again.
• If your microphone itself does not have a USB connector and you are using a USB audio adaptor or audio interface, then you must look for the USB device name and not the microphone name.
• If your microphone is connected via a standard small (3.5mm) microphone jack instead of USB, you will not see the microphone name, but depending on the system you might see something that indicates an external audio device; see what it is called in your computers control panel or system preferences and look for that in the Settings dropdown.
• In some cases on Mac OS X systems, microphones connected with a 3.5mm jack will still show up as "Built-in Microphone." This make it very hard to tell if you are really using the microphone and can lead to problems. We recommend using a USB adaptor or microphone.

What to Do About Clipping

Clipping is when brief portions of the recording are so loud that they cannot be accurately captured digitally. When this happens, it is necessary to re-record the utterance. One of the most common causes of occasional clipping is having the microphone positioned directly or nearly directly in front of your lips so that it captures the bursts of air associated with "plosive" consonants like /p/, /t/, /k/, and others. If that is the cause of the clipping, the best solution is to move the microphone down so that it is about parallel with your chin. In that position it will be less susceptible to these bursts of air.

There are two general types of noise that might be a problem for recording:

• External noise - Noise from appliances (fans, A/C units, heaters), other people and pets, reverberant rooms, open windows, etc.
• Internal noise - Created by your computer and audio hardware such as electrical hum or static.

For internal noise, if you are working on a laptop that is plugged in, you may be able to reduce electrical hum by unplugging the laptop and running on battery. If you are not working on a laptop, or if you are and it is already unplugged, try moving any fluorescent lighting away from your work area. If neither of these solve the problem, it is possible that you will need a different Microphone, or even a different computer to do the recording.

Additional Hints for Successful Recording

• You must be seated in a quiet non-reverberant room to do the recording. A small room is better. The more carpeting, drapes, and overstuffed furniture the better.
• A trick that radio professionals sometimes use in unfavorable recording environments is to drape a blanket or quilt over the person with the microphone to help damp out other sounds and echoes. You can cover yourself and your monitor or laptop with a quilt to block out the room sounds as long as your monitor or laptop are not also very noisy.
• Use a good quality head-mounted microphone that connects to your computer via a USB audio interface or professional-grade sound card.
• Speak in a way that is natural and comfortable for you. Use the spoken prompt to understand how you should say each utterance. Sometimes we are looking for a simple statement or question. Other times, we may want you to express a particular sentiment or interpretation by empahsizing some words more than others.
• Recording stops as soon as the app thinks you are done speaking, so do not insert long pauses within the sentence.
• Each sentence is uploaded to our server as soon as it is recorded. If your outgoing internet service is slow, there can be a longish pause (several seconds) before the next sentence is presented.
• Our server keeps track of where you are in the recording sequence, so you can leave off recording when you are tired or need to take a break.
• Try to record at the same time of day every day until you are finished.

Privacy

The speech you record will be stored securely on our servers and may be used internally to help evaluate and improve our recording and synthetic voice building procedures. We will not share your name, or any personal information with anyone outside our project without your written permission. We will share examples of your speech or synthetic voice only if you gave us permission for this when you registered.