This page was originally written in October 2002, and has been substantially updated to only cover recording bells for tonal analysis, not recordings for general purposes. Also, audio technology has moved a long way since 2002!
Andrew Wilby suggested I wrote the original paper. It included extracts from messages originally published on the Bell Historian’s chat list. The following people helped with ideas and experience: Andrew Aspland, Alan Birney, George Dawson, Stephen Ivin, Dave Kelly, Nigel Taylor and Michael Wilby. Shahram Taherzadeh of the Open University helped me with microphone calibration.
There was a useful article on this subject in the Ringing World of 24th April 1987, pages 379 and 380.
To do tonal analysis of bells (with Wavanal), recordings are needed in a digital format, either .wav or mp3. Either are equally suitable. If any manipulation, editing or conversion of recordings is needed I strongly recommend Audacity which is a free and very comprehensive sound editor and recorder.
Many devices can be used to record bells. Examples from the past include reel-to-reel tape recorders and cassette recorders. More modern devices include video-camcorders, laptops using the built-in microphone or an external microphone, or mobile phones or tablets.
Recording direct to a PC or laptop saves a step in format conversion. Using a mobile phone or tablet has the advantage of speed and portability. I routinely use either a laptop with built-in microphone or a mobile phone for recordings. For the very best quality I use the laptop with a USB pre-amp and a separate condenser microphone but this is overkill for recordings for tonal analysis.
The problems encountered in practice are:
- recorder speed stability
- automatic level recording or noise suppression
- microphone quality.
Frequency stability of the recording device is important if the recording is to be used for tonal analysis or pitch determination. Frequency stability is not an issue with digital recorders, and nor does it seem to be a problem with camcorders. Cassette recorders, unless they are very good quality, are poor. Two problems are experienced; devices that record at a different speed to playback, and devices where the recording speed fluctuates during the recording. It is possible to get around frequency stability problems with the use of tuning forks or reference handbells, but this is somewhat inconvenient.
For very accurate measurements, I calibrate the devices I use (either a laptop or a mobile phone) using the NIST standard tones as explained here. However, over a number of years and many calibrations I have not seen a discrepancy, the devices I use are stable to better than 0.05Hz.
Recording devices with automatic level control should be avoided like the plague. Because bells should be recorded in a quiet environment, and are very loud, recordings taken with automatic level control usually consist of loud background hiss interrupted by the fainter sound of the bell or bells. Digital recording devices and camcorders seem not to have auto-level control. More subtle but also frustrating is the noise suppression in laptops. This causes the recording level to drop to zero as the sound gets quieter, which can clip the end of a bell sound or worst case (with quiet bells) cause nothing to be recorded. There is a setting on PCs allowing noise suppression to be turned off.
If bells are being recorded for tonal analysis there is no need for a separate microphone, the microphones in modern laptops and mobile phones work fine.
Microphone quality only matters if the microphone is situated very close to the bells (in the bell chamber). Bells are very loud when struck hard or ring full circle and microphones can easily overload. When recording bells close up, I use condenser microphones which give distortion-free results only a few feet from bells being rung full circle. I use a pair of AKG C1000S microphones which I bought many years ago but any modern condenser microphone with a good maximum SPL (sound pressure level) should work fine. I have no personal experience of dynamic microphones but their frequency and transient response suggests they would be less good. There are modern condenser microphones with a USB interface allowing them to be plugged direct into the laptop. Microphones using XLR cables will need a pre-amp with phantom power, these are available with a USB interface, again allowing direct connection to the laptop.
If recording for tonal analysis, bells should be recorded singly, not rung in changes or chimed together. It is sometimes possible to analyse the individual bells from a recording of changes but it is very time-consuming (it can take many hours) and the results are never certain.
The standard approach I use is to chime or ring each bell individually with a ten-second gap between each bell – to give the sound of each bell time to die away, and to give a reasonable length of recording for analysis.
The most accurate measurements can be when the bells are stationary and hit hard with the clapper. A hard blow is needed to stimulate the upper partials, a gentle tap means the higher frequency partials will be quiet and hard to detect. Next best, and not requiring access to the bell chamber, is to record each bell chimed while down. It is practical to use recordings of bells rung full circle, but the frequency measurements will be less accurate due to the doppler shift produced by the moving bell.
Recording bells which are not hung, e.g. standing on the floor or on timbers, is not satisfactory. Some of the partials can be identified in recordings but the rim partials are generally supressed, and if they can be identified, their frequencies may be changed by the physical contact with the floor.
Tonal analysis of cracked bells is never satisfactory, unless the crack is in an area of the bell that doesn’t change its tone.
When recording installations with a large number of bells in the bell chamber, it is very useful to keep track of which bells have been recorded. More than once (for example, when recording the carillon at Manchester Town Hall) I have returned home to find that one bell has been missed!
Bells are very loud indeed, as we all know. I have found that a single clapper blow recorded close by the bell can easily overload the various recorders and microphones I use. There are two ways to get around this. First, distance always helps – recording from another floor, further from the tower etc. Learn the characteristics of your recorder and how close you can get without overload. The second cure is to always do a level check before taking a final recording, and either move further away or turn down the record level to get satisfactory results.
Just because the recording is not audibly distorted does not mean that you are free of overload – looking at the file with a sound-file editor will tell you if clipping is occurring. Aim for no more than 80 or 90% of full scale to be sure. If it is not possible to view the waveform in this way and your recording device has a meter, do not allow the record level to go above 5 or 10 dB below the maximum. The brief but high intensity partials produced by the clapper blow die away before the meter can register them. Aim low – provided the recording is reasonably noise-free, quiet recordings can be analysed, but there is no way to remove overload.
Digital recording formats
With advances in technology, file size and recording format is no longer the issue it was 20 years ago. I have 9,706 audio and video recordings of bells which occupy only 17.1Gb on disc. The Windows installation on my PC alone occupies 7Gb! So optimising recording format to save space on disc is no longer a concern.
As regards sampling rate in sound files, a theorem known as the Nyquist Theorem, grossly summarized, says that the sample rate should be double the maximum frequency present in the sound. In practice I find with recordings you need to go higher than that. For the best results, sample at 44,100 samples per second. Recording at 22,050 runs the risk of losing frequencies somewhat below 10 kHz but may be acceptable if the original recording is bandwidth limited. recordings I have done at 11,025 sample per second miss some high partials in small bells.
As regards sample size, 8 bit data gives disappointing results, 32 bit data is complete overkill and is not universally supported. Use 16 bit. Except for recordings in exceptional circumstances (e.g. recording a swinging bell close up) I do not believe stereo adds any value to recordings of bells. Save the disc space and use mono!
On a PC, the recording formats generally encountered are .wav and .mp3. Wav files are uncompressed and therefore larger. Extensive experience with MP3 files shows that they cause no issues at all in tonal analysis of bells. Many other audio and video formats will be created from other devices but in practice all can be converted to .wav or .mp3 using Audacity. If the original file was compressed in any way, convert it to .mp3, there is no additional benefit to the disc space taken by a .wav file.
I take a lot of bell recordings from youtube, which is a fantastic source for bell recordings, especially for bells in mainland Europe. I use an online converter to download the audio as an mp3 file. I won’t recommend a particular conversion site here as they regularly go offline or are overtaken with adverts. Google for ‘online youtube to mp3’ to find many links.