Improving RealAudio Input Files

Contents:

1. What To Expect
2. What To Do First
   • Use a good original source
   • Set your input levels correctly
   • Use good equipment
   • Select appropriate codec
3. Signal Processing
   • Correct DC offset
   • Noise Gating (or Downward Expansion)
   • Compression
   • Equalization
   • Normalization
4. Final Notes
5. Other RealAudio Resources
6. Software Package Specific Hints

What to Expect

Approximate file sizes for one minute sound files

.wav file sampled at 44kHz, 16 bit Stereo	10.1 Megabytes
=>encoded into RealAudio Dual ISDN Mono/Stereo	586 Kilobytes
.wav files sampled at 22KHz, 16 bit Stereo	5 Megabytes
=>encoded into RealAudio Single ISDN Mono/Stereo	296 Kilobytes
.wav file sampled at 8kHz, 16 bit Mono	938 Kilobytes
=>encoded into RealAudio 3.0 28.8 Mono	117 Kilobytes
=>encoded into RealAudio 2.0 28.8	113 Kilobytes
=>encoded into RealAudio 1.0 14.4	60 Kilobytes
New! RealAudio 4.0 8.5Kbps	64 Kilobytes
New! RealAudio 4.0 6.5Kbps	42 Kilobytes

The only way to achieve such compression is by leaving out some of the soundfile information. The RealAudio encoder does this by identifying which parts of the soundfile are most important (more on this later). No matter how carefully a file is pre-processed, though, the 60K RealAudio file will never have as great a frequency response or dynamic range as the original 10 Megabyte file.

Low bit-rate codecs	You should expect sound quality equal to a clear, strong AM radio station, with slightly reduced frequency and dynamic range
Medium bitrate codecs RA 2.0 28.8, 3.0 28.8	You should expect sound quality equal to a clear FM radio station, albeit in mono and with slightly reduced frequency and dynamic range
High bitrate codecs ISDN	You should expect near-CD quality, with a slight reduction in frequency and dynamic range.

What To Do First

1. Use a Good Original Source.

   A high-quality audio source is probably the single most important variable in determining your final audio quality. At RealNetworks, we start with satellite signals, Compact Discs, or Digital Audio Tapes (DAT). When creating sounds from scratch, we use professional-quality microphones. You can make sound files from low-quality analog cassettes, tiny condenser microphones, or anything else -- but the hiss and distortion in the resulting sound file will have a substantially adverse effect on clarity after the file is encoded into the RealAudio format.

   You should always encode from 16-bit (not 8-bit or mu-law) sound files. We also recommend digitizing at a minimum 22050 Hz sampling rate. If you plan on offering high bitrate RealAudio streams, you may even want to digitize at 32KHz or even 44.1KHz. In practice you will have to balance your quality demands with storage space available; 44.1KHz stereo files eat up drive space!

2. Set Your Input Levels Correctly.

   Setting correct levels is absolutely crucial. All audio equipment has an inherent signal to noise ratio. This is the ratio between the loudest possible sound it is able to reproduce (without distortion) to the noise inherent in the equipment. All audio equipment has a noise floor; there is no such thing as an absolutely silent piece of equipment. To take full advantage of your equipment, you want to work with the loudest sounds possible.

   When creating your original sound file, the input level should be set to use the full range of available amplitude without distortion. This is known as &clipping,& and is audible as a high frequency crackling noise.

   When digitizing with your sound card, first do several test runs and adjust your input level so the input approaches but does not exceed the maximum level. You can adjust this on the mixer page of your sound card utilities. Look for the "Input Levels" or "Recording Levels" option. Most mixer pages have some sort of visual display where you can see how much sound is coming in. Make sure there are no peaks above maximum. These are generally indicated by a red light somewhere. Be conservative with your levels; you never know when someone will get excited and speak much louder, or when a big play at a sports event will make a crowd roar. Differences in volume levels can be evened out later (see Compression).

   Sound files that do not use the full amplitude range will produce poor-quality RealAudio files. If the amplitude range of an existing file is too low, you can use your audio editor's "Increase Amplitude" or "Increase Volume" command to adjust the range before encoding the file. Most sound editors have a "Normalize" function that will maximize your levels automatically (see Normalization).

   Note, however, that better quality will be achieved if the levels are set correctly at the time of recording. The good news is that once you set your input levels correctly, they generally will not need to be reset. If you are reasonably consistent with your recording practices, you will save yourself a lot of trouble in the long run.

3. Use High-Quality Equipment

   High-quality equipment will produce better results and save you a lot of headaches in the long run. Every piece of equipment in the audio chain, from the microphone to the soundcard to the software, will have an effect on your RealAudio files. If you intend to be a commercial RealAudio content provider, you should invest in professional quality audio equipment. This need not be a crippling investment, but it does mean you will have to purchase from a professional recording equipment dealer, not your local computer/hi-fi/gadget store.

   For an example, see what we use at RealNetworks.

4. Select Appropriate RealAudio Codec

   With the release of RealAudio 4.0 in February 1997, there are now a wealth of codecs for you to choose from when producing audio content for your website. Take a few minutes to think about what sort of programming you are delivering, what your target audience connectivity is and what your bandwidth limitations are.

   Let's take a look at your options:

   • ISDN codecs
     • Suitable for all content, especially intranet based material.
   • 3.0 28.8 codecs
     • The majority of web users still connect at this rate. We have provided three codecs at this rate, each specifically tailored to a particular type of content. The "full" codec has the best frequency response and is well suited for classical music. The "medium" codec has a slightly reduced frequency range, but is better suited for rock/pop music. The "narrow" codec has the narrowest frequency response but has less artifacts. It is well suited for news, talk, and interviews.
   • 2.0 28.8 codec
     • Our original music grade codec. It has been superceeded by the 3.0 28.8 codecs, but some sites choose to continue supporting this codec because of the large RealAudio 2.0 listener base who have yet to upgrade.
   • 4.0 Low bit rate codecs
     • These new codecs were intended to be used with video encodings, but we discovered the 8.5Kbps codec outperforms RealAudio 1.0. For video purposes, we recommend the 6.5 Kbps for voice grade applications, and the 8.5 Kbps for music grade applications. In general, the 8.5Kbps is an outstanding performer in the low bit rate class.
   • 1.0 14.4 codec
     • The original RealAudio codec. Extremely robust for voice grade applications. In theory it has been superceeded by the 8.5Kbps codec, but the huge number of listeners who have yet to upgrade means a lot of sites will continue to support this codec.

Signal Processing

The RealAudio Encoder makes files smaller by making educated guesses about what is most important in the soundfile. It knows how much room there is in the destination stream and fills that available bandwidth with as much sound information as it can. Any sound information that does not fit is lost. We can help the encoder with its task by emphasizing the most important parts of the recording.

What tools are available to us? Excluding special effects, all signal processing essentially breaks down into two categories: gain control (or dynamics processing) which includes noise gating, compression, and limiting; and equalization (EQ). With dynamics processing, we can adjust the volume levels of the input signal. EQ allows us to emphasize certain frequencies and de-emphasize the ones we don't want, that may be extraneous to the program material or may be clouding up the original signal.

With the higher bitrate codecs, you'll find that little or no signal processing is necessary. The encoded versions are more faithful to the original recordings. However, with the lower bitrates some signal processing is generally helpful.

There are several straight-forward processes that you can use to improve your audio signals, most notably noise gating (or expansion), compression, equalization, and normalization. Additionally you should always check for and correct DC offset, which can wreak havoc at lower bitrates.

These processes are discussed in terms of the hardware used in recording studios; however, many of these same features are also available in software form for Macintosh and PC platforms. Depending on the package, you may find some or all of the options supported.

Note: Many software programs will not process audio files in real time. Those that do will have trouble processing with the RealAudio Encoder running at the same time. In addition, on most platforms you will run into filesharing problems. If you are thinking of live broadcasting, you will have to consider getting hardware versions of the tools described below.

1. Correcting DC Offset

   Sometimes when files are digitized, something known as DC offset creeps in. This is when the digitized waveform is not correctly centered around the 0 volts axis. Most of this is due to improper grounding of soundcards. Some soundcards are worse than others; to see how bad your soundcard is try recording silence. You should in theory see nothing in you waveform window, but you'll probably see a flat line just slightly above or below the 0 volts axis. This is DC offset.

   This can wreak havoc when you attempt to process your waveform, and can add a low rumbling sound to the encoded file. Luckily most editors have a built in facility to take care of this. Some call it 'Centering the Wave' and are automatic; others allow you to adjust DC offset manually (+/-). In this case you'll have to find out precisely what your DC offset is by running a &statistics& command or something similar. Then you'll have to correct it. For instance, if your average DC offset is 45 you'll want to offset the wave by -45. Please consult your software manual and follow the instructions.

   Obviously if you are doing a live broadcast, you will have to live with whatever DC offset you have. Proper balanced wiring between all your audio components will help minimize this as well as any ground loops.

2. Noise Gating (or Expansion)

   Noise Gating, or downward expansion, eliminates unwanted background noise which becomes audible during pauses in the audio (e.g. when an announcer pauses, or there is a gap between programs). Signals above a certain volume level are left alone, but below this level the signal is turned down or even off, depending on how heavy the gating or expansion is. Setting up a noise gate or expander is straightforward. Most budget compressors have a noise gate built in.

   To use noise gating, set the threshold control so that the gating or expansion occurs when there is no desired audio, but not so high that the beginnings of words or music that you want to hear are chopped off. It takes a bit of time, but remember to err on the side of caution just in case the next person in the program has a softer voice.

   If your gate or expander has a range control, set this to around 5 - 10 dB. This means it will turn down the "noise" sections a little, but not turn them off altogether. That way you'll hear if the gate is cutting something off that you want to hear, and you can then readjust the threshold setting accordingly.

3. Compression/Limiting
   

   (Note that this process -- audio compression -- is unrelated to file size compression, mentioned earlier.)

   As discussed earlier, all audio equipment has inherent noise, and to take full advantage of a piece of equipment's signal to noise ratio, we want to feed it as loud a sound as possible.

   In addition, one side effect of RealAudio encoding can be artifacts - sounds that were not there before encoding. Some artifacts can be heard as rumbling or distortion in the signal. This type of artifact appears at a relatively constant low level, whether the original soundfile was loud or quiet. Louder files tend to &mask& these quiet artifacts.

   So for these two reasons, we recommend feeding the encoder a loud signal. However, we are limited by the loudest section of the file being encoded. If we could turn down the loudest section, we could turn the overall volume of the soundfile up. A compressor helps us accomplish this.

   

   Compression reduces the difference between the loudest and quietest sections of the incoming signal. Sections that exceed a user-defined threshold are turned down.

   Compression - Graphic Examples

   The same incoming signal with two different thresholds.

   

   

   Note the different results. The signal on the left can be turned up substantially.

   
   A compressor allows us to encode louder files by giving us more "headroom."
   

   How much compression should you use?

   The exact settings will be determined by experience and by referring to the manual that comes with your equipment or software. Here are some rough guidelines which should be helpful in most situations:

   Low bitrate codecs	Use moderate to extreme compression (4:1 to 10:1). This guarantees the resulting signal will be loud enough to mask artifacts which may appear. When encoding music, always make sure the result is satisfactory.
   Medium bitrate codecs (28.8)	With the RealAudio 3.0 codecs, the dynamic range is greatly increased, and the artifacts are greatly reduced so the need for compression is not as great. A compression ratio of between 2:1 and 4:1 is generally more than enough. With music, compression effects are more readily apparent. Always check your results.
   High bitrate codecs (ISDN)	At these bitrates, compression may not even be necessary. It might be more of a precaution against clipping if you are sreaming a live feed, or just a mild compression to get a slightly 'hotter' (louder) signal.

4. Equalization

   Equalization (or EQ) changes the tone of the incoming signal just as you can on your home stereo or car radio. This is done by &boosting& (turning up) or &cutting& (turning down) certain frequencies. Using EQ, we can emphasize frequencies that we want (where the important content is) and cut frequencies where noise or unwanted sound is. By doing this, we can give the encoder a big hint about which sound information to keep.

   In addition, some codecs do not have a flat frequency response. In other words, certain frequencies may not be as loud after encoding as in the original. We can compensate for this with EQ.

   Low Bitrates

   • At low bitrates, the RealAudio encoder discards the high end, or 'treble' information. This can make files sound dull. To compensate for this, it helps to boost the middle frequencies or &midrange.& This will also make speech sound more intelligible.

     Most mixing boards will have a midrange EQ knob. Sometimes you can choose which frequency to boost, other times this is preset at the factory. If not, or if you are using a graphic equalizer or audio processing software, you want to boost at around 2.5KHz.

     If your equipment does not have a mid frequencies EQ knob, you can obtain a similar result by turning the low and high EQ knobs down and then turning the overall volume back up (note, though, that this is not as effective as boosting the mids, which attacks the problem at its source).

     The amount that you should turn up the midrange depends on your EQ equipment and source file. A little experimentation is necessary. Try adding some mids to a short section of a piece to be encoded and check it with the RealAudio Player. If it is a bit muddy or hard to understand, try adding a little more. You can keep going until the knob won't turn anymore, or until the result starts to sound too harsh.

   Medium Bitrates

   • Again, with the new 3.0 codecs, the use of EQ is a more artistic process. Because the reproduction is more faithful, nowhere near as much preventative EQ is necessary. A bit of midrange boost is probably still the order of the day to make up for the lack of treble information.

   High Bitrates

   • At high bitrates, there is no need for EQ as long as the original source was of sufficient quality. You may need to use EQ as a tool to get rid of noise, or to brighten up a dull recording, but that is an aesthetic judgement to be made with care. Because of the near CD quality at higher bitrates, any change you make to the original will be heard in the final encoding. Make sure you are improving the source, not just changing it.
   
   
5. Normalization

   Normalization is a process included in most audio recording software whereby the computer calculates exactly how much it can turn up the volume of a file without distortion. Because we always want to feed the encoder the loudest files possible, this is a very handy function. This is why you can afford to be fairly conservative with your recording input levels, and then let your program's normalization function take care of the rest. There are a few things to take into consideration:

   1. Normalization should be the last thing you do. If you normalize your file, and then ADD some EQ, you'll end up with distortion. (If you add after you've maximized you will exceed)

   2. The RealAudio encoder sometimes chokes on peaks when they are too close to the maximum. Normalizing to 95% of maximum usually avoids this. If your Normalization option will not let you specify a percentage, simply turn down the overall volume after you normalize using the &volume& or &amplify& option in your audio software.

Final Notes

Equipment to do this need not be outlandishly expensive. Think about what it is you hope to achieve, decide on a budget, and consult with your local recording studio equipment dealer. They should be able to come up with a working solution, be it hardware, software, or a combination of both. Most importantly, make sure that they are going to send someone out to help you install your system. What would take a knowledgeable engineer half a day to set up could frustrate a novice for weeks.

One final note: some people's voices are more suited to recording than others. Professional announcers are paid not only for their reading abilities, but also for their pacing, dynamics, inflection, and most importantly vocal tone. Bear this in mind when choosing the person behind the microphone.

Other RealAudio Resources