Basically, there are three types of noise that are removed during audio restoration.
- Impulse noise, which sound like clicks, pops, and crackles that are caused by vinyl imperfections, scratches, and contaminates in record grooves.
- Narrowband noise, which is represented by specific frequencies. Examples would be; 1) 60 Hz noise (and its harmonics) that are caused by poor shielding and grounding and 2) lower frequency rumbles that are sourced by the turntable.
- Broadband noise, which is a wide range of random frequencies (e.g. hiss and static)
The good news is that after completing the music restoration process, a majority of these noise sources will be virtually removed and the audio quality of the music, even starting with a moderately worn record, may sound as good or even better than the original record. You’ll find that for most records, music restoration process will take 5 to10 minutes. In other extreme cases, some additional effort and time will be required to achieve desired results. Any noise removal technique will also have some impact upon the original music content so it is important to apply just enough noise restoration techniques to achieve an acceptable music quality. Being too aggressive when applying the noise removal tools can remove real music elements and even worse, introduce new, undesirable sound artifacts.
In addition to noise removal, two additional processing steps are required during the music restoration. The first is volume normalization and the second is converting the processed music data from a 24-bit representation to a 16-bit format.
To achieve the best results, the music restoration processing sequence should be:
- Trim recorded data.
- Eliminate impulse noise.
- Filter narrowband noise.
- Normalize volume of music file.
- Remove broadband noise.
- Convert file to 16-bit format.
Prior to starting noise removal processing, silent gaps and extraneous noise should be removed from the recorded music file using the Sound Forge “cut” tool. Although the silent gaps could be removed at any time during the audio processing sequence, the large transients that are caused when the turntable stylus is queued onto the record need to be removed so they don’t interfere with noise removal processing. At least one 2 second silent period should be retained, usually at the beginning or end of the recorded file, which will be used to support broadband noise removal processing.
The Sound Forge Noise Reduction DirectX plug-in includes four different plug-in tools, two of which are used for music restoration. The first one that is used is the Click and Crackle Removal. The Sound Forge tools are very powerful, but can be somewhat complicated for new users. But what’s nice about these tools is that they come with pre-defined configurations that have been optimized for different types of audio processing. For example, when using the Click and Crackle Removal tool, a “default for vinyl recordings” processing configuration can be selected, which does an excellent job on most records. So after selecting the “default for vinyl” settings, select the entire file for processing (i.e. Ctrl A) and start the Click and Crackle Removal processing. After the processing is completed, play regions of the file and visually review the file waveform to verify that all of the worst impulse noise artifacts have been removed. The tool may not remove some larger clicks. Rather that applying the Click and Crackle Removal tool again to remove a few large remnants, it is usually better to manually eliminate these noise artifacts using the Sound Forge pencil tool. Using this tool, a large spike can be eliminated by simply drawing a line to replace the spike. Be sure to manually correct clicks on both right and left channels. If there are a large number of remaining clicks that need to be removed, it may be more effective to run the Click and Crackle Removal tool using the same settings. Running this tool a second time may have a slight adverse affect on the music, so be sure to listen to the music before and after applying additional processing to determine what processing provides the best results. Note that using the default “vinyl” settings also applies a filter to eliminate low frequency “rumble” noise artifacts. If this tool is applied a second time, be sure to unselect the “Remove low frequency rumble” option to avoid over-filtering of lower frequencies.
Since the default settings for the Click and Crackle Removal tool includes rumble filtering, no additional narrowband filtering is required. The only other noise filtering that could be considered is a hum filter to attenuate 60 Hz noise and higher frequency harmonics that are caused by grounding and shielding problems. I’ve evaluated different types of hum filters to assess the potential audio benefits. But after experimenting with different types of music and reviewing the spectral effects of the filtering (using the spectrum analysis tool in Sound Forge), I came to the conclusion that if the grounding and shielding is done properly, this filtering step is not necessary.
Before using the Noise Reduction plug-in it's best to normalize the music volume first, since the process of normalization will increase the amplitude of some small noise components. If noise reduction is done after normalization, the broadband noise artifacts are more effectively removed.
When music is recorded from vinyl, the volume during playback may vary considerably since the original recorded volume and the volume setting during digitizing is going to be different for each record. Ideally, all tracks in the music library should playback at the same volume for a given volume setting. To achieve this goal, the volume of the recorded music must be normalized. There are generally two types of normalization; peak amplitude normalization and RMS normalization. The first method identifies the maximum audio level, then adjusts the value of this point to a value of 0 dB, and then applies the same gain to all other digitized points. Since only one point is used as a reference value to calculate the normalization adjustment, one short, loud sound can skew the results of this type of normalization. A more effective normalization scheme calculates the average music volume, then makes a volume adjustment based this calculated average and a desired maximum volume level. This is equivalent to RMS normalization, which is supported by Sound Forge. The Sound Forge normalization function also supports a large number of settings. However, Sound Forge again provides a preconfigured setup for that has been optimized for music normalization (RMS normalization at –16 dB), which should be used. After normalization is completed, the amplitude of the audio waveforms should appear larger and the music will generally sound louder when played. But more importantly, if done to all your recordings, the loudness of all your recordings will be relatively the same during playback.
After normalization is completed, the Sound Forge Noise Reduction tool should be used to remove any remaining broadband noise. This noise reduction technique uses a short silent recorded interval as a noise print reference. Any recorded information in this “silent period” is considered background noise, which is then removed from the entire recorded music file. Selecting one of the “default auto capture” modes provides an optimal setup for normal noise removal. If after processing, the file still contains significant noise artifacts in the silent recorded periods, you can run the Noise Reduction tool a second time to further reduce the broadband noise components. Note that at any time during audio processing, the “undo” function can be used to quickly compare the audio quality before and after a processing step.
The final processing step is to convert the 24-bit file format to a 16-bit format using the Sound Forge Bit Depth Converter function. When using this function, the “Convert to 16-bit (dither and noise shaping)” preset should be used. After the conversion is complete, save the file in a Microsoft WAV format.
