Speech Recording

Anyone can record speech, but there are some considerations that are different to those when recording music. Attention to a few details can improve the quality of speech recording.

NOTE: Some contents of this page have been integrated elsewhere under Recording Tips, and reference to this page has been removed from Recording Tips. The page may well be listed for deletion in the near future if all content is merged elsewhere.

Related pages:

Our complete set of Recording Tips

Technical Issues

Dynamic Range

One feature of recording people speaking is uncertainty of recording level. Speakers vary in volume, and may not be aware of the best microphone techniques so for example may stand in different positions relative to the microphone. In some cases, such as meetings and conference recording, there may also be remote participants who are being heard through a radio or television receiver. The result is wide recording level variation.

Rather than record at the final bit depth wanted (let's say 8 bits), with digital recording one can record at greater bit depth and set the recording level relatively low (say 10 dB to 20 dB below the -0 dB distortion level). This retains plenty of dynamic range but avoids the risk of speakers who are louder than others creating clipping, which will result in unpleasant sound quality.

When recording with a computer, the greater the bit depth you record at, the more CPU  time and disc space will be used. See Bit Depth for more explanation.

Sample Rates

There are also pros and cons about recording at different sample rates. The sample rate of the recording determines the highest frequencies that can be captured. Generally, lower sample rates are acceptable in speech recording where they are not in music, because voices (especially male) have a lower upper range of fundamental frequencies than instruments. Also, by the nature of the different sounds made when speaking and singing, it's less important for quality reasons to capture the higher overtone frequencies in speech. In any case, the higher the sample rate you do record at, the more CPU time and disc space will be used.

Multi Channel Recording

Where speakers don't stand close to the microphone, multi-channel recording helps to keep all speakers above the room noise level, and clearly audible.

For meetings it may be useful to place several microphones around the room, recording each microphone on its own channel. The multiple channels can be mixed down to mono later, selecting for each speaker whichever channel gives the highest ratio of speech to room noise. When post-processing, simply choose one channel for each speaker, mute the others, then mix down.

Once again, more channels mean greatly increased CPU use and greater use of disc space. It's important to test the hardware in multi-channel mode in advance, as running out of CPU capacity could cause recordings to have drop-outs or fail completely.

Where simplicity is required, using only two microphones in different positions can still significantly improve the end result.

Reliability

The generally more stable nature of Linux or Unix operating systems may mean a reduced chance of a recording failure if you record with these systems rather than with Windows (other things being equal).

Up-to-date sound card drivers specific to your hardware are more reliable than generic drivers when recording. Be prepared so you can quickly reinstall sound drivers between events if necessary.

Shutting down un-necessary applications and processes so that the recording has most of the available CPU to itself (for example with tools like Winsolo  or Enditall  for Windows), is important - especially on slower and older machines with less RAM. Don't make text notes on the PC that is recording, or do anything but record with it. Such actions are likely to cause recording skips. A separate laptop is good for that.

Consider making a checklist for any important recording. You may want to do a last minute check that you've got power settings set to always on, screensaver off, levels set right and so on before you record. The Wiki has more tips on Managing Computer Resources and Drivers and further General Recording Tips that you should read before commencing an important recording.

No PC hardware or software can be relied on with absolute certainty. A reliable backup recorder is very much recommended. An mp3 recorder is a good back-up recorder if it has sufficient capacity and works with external microphones.

Tape-based recorders may not deliver good sound quality or be able to record uninterrupted for long periods, but a recording made with them is less likely to fail than one made with a computer, and after processing with Audacity (given this is speech rather than music) the results should be acceptable.

Backing up the computer audio files as soon as possible is wise, and keeping the original and back-up files physically separate avoids loss in case of theft, breakage or spillage. Giving someone else a copy of the files each day reduces the chance of data loss.

Microphones

This section is aimed at home users rather than professionals, who will know what they're doing with microphones.

Never use a built-in microphone that comes with a laptop, MP3 recorder or tape deck. Such microphones pick up lots of noise from the device's drive or from the deck motors or tape.

While microphones are usually set on stands for formal events, for meetings of a handful of people held round a table, microphones on the table may be sufficient. Always place the microphone on something soft and squashy so that sounds and vibrations transmitted through the table - of which there are usually many - are not picked up directly. The microphone lead itself should sit on the squashy item before it reaches the table, as some sound and vibration can be passed up a short length of cable.

The squashy items should be stable however; sponges fail in this respect! Folded clothing works fine, and the informal appearance helps put speakers a bit more at ease.

Alternatively, microphones hung overhead avoid vibration and disturbance.

If using cheap home use microphones, I can recommend condenser type rather than moving coil, as quality is relatively consistent with these, and cost is minimal. Moving coil microphones can be more variable.

If using leftovers on zero budget, avoid crystal microphones! These read open circuit on a multimeter, and distort horribly. Distortion causes intermodulation , which makes it difficult to remove noise effectively in Audacity.

Distribution of recordings

Don't overlook the reality that a lot of people are still using audio equipment and computers which are 10 or 15 years old. Many may not even have a computer or MP3 player. So be prepared to make your recordings available on audio CDs for those who don't have computers or a DVD player that can play longer "data CDs". For distribution as a computer audio file, stick to MP3 as it's a universal format that virtually any computer will be able to play irrespective of age or operating system.

If you want to be thorough you could even include a lightweight portable freeware player that runs on win 95 on a data CD, and plays whatever format you're distributing.

Human Issues

Speaking Technique

There are plenty of web pages on microphone technique such as this one . Many concentrate on singing technique but similar principles apply. If you get a set of speakers willing to pay any attention to such things, lucky you!

Generally the less attention speakers pay to microphones the better their talk, and a way to minimise awareness is to not even mention the subject.

Hidden microphones can put speakers more at ease. They still know the microphone is there, but not being repeatedly reminded of it helps.

A basic way to create a hidden microphone is to cover a black-coloured microphone and its visible section of lead with a layer of lightweight open weave clothing.

Reluctant speakers can sometimes be enticed forward by informing them that they get to review their recording and can say no to its distribution if they wish. This approach is only suitable for some situations of course, but it does elicit more spoken material. Reluctant speakers tend to fear things that don't usually happen, and usually say yes to distribution afterwards.