GB2403386A

GB2403386A - Method and apparatus for signal processing

Info

Publication number: GB2403386A
Application number: GB0314417A
Authority: GB
Inventors: Anthony Paul Webster; David Anthony Betts
Original assignee: Cedar Audio Ltd
Current assignee: Cedar Audio Ltd
Priority date: 2003-06-20
Filing date: 2003-06-20
Publication date: 2004-12-29
Also published as: GB0314417D0

Abstract

A method of equalising an audio signal preserves a given loudness measurement function of a nominal input signal. The nominal input signal may be generated by a pink noise generator 2, and the loudness measurement function may be a weighted power function such as standard "A", "B", "C" or "D" weighting curves. The audio signal may be equalised by a conventional equaliser 3 and 4, with an additional gain. The additional gain 8, may be calculated by measuring the loudness of the nominal input signal before (forming the nominal input loudness) and after (forming the nominal output loudness) it is passed through the equaliser, with the same equaliser settings as those being applied to the audio signal, 5. Both the nominal input and output loudness are calculated according to the same function, steps 6 and 7. The additional gain is applied in step 9 resulting in the desired audio output signal 10.

Description

- 2403386 Method and Apparatus for Audio Signal Processing

Field of Invention

This invention concerns methods for the equalization of audio signals.

Background to the Invention

Equalisers are used in audio signal processing to boost or attenuate signals at certain frequencies. Their original use was to compensate for the non-ideal characteristics of a transmission channel. For example, analogue telephone circuits suffer greater signal loss at high frequencies than at low frequencies, so an equaliser is used to boost the high frequency relative to the low, and thereby to "equalise" the frequency response of the channel.

Subsequently equalisers became commonly used to adjust tonal balance to suit the listener's taste; witness the graphic equalisers and the bass and treble tone controls on domestic audio equipment. Sound studios have more sophisticated equalization tools available to them: parametric equalisers, low shelves, high shelves, band pass and notch filters are all types of equaliser that are frequently used in the production of audio material.

However, one problem with using equalisers for adjusting tonal balance is that an equaliser generally changes the perceived volume as well as the tonal balance, and the audio engineer is often fooled into thinking that one particular setting sounds better than another simply because it is louder.

The loudness of an audio signal is an objective measure of the ear's subjective perception of how "loud" that sound is. It is difficult to quantify because of the complex behaviour of the human auditory system. It is related to the acoustic wave intensity, this being a measure of the rate at which energy is transported by an acoustic wave. In the case where the acoustic wave is generated by an electrical signal fed to a loudspeaker, the intensity can be shown to be closely proportional to the electrical power delivered to the loudspeaker. Thus, a measurement of the electrical signal power forms a useful first approximation to the loudness.

In 1933 Fletcher and Munson studied the loudness of tones of constant intensity as they varied in frequency. Their findings (and those of subsequent research) lead to more sophisticated methods of measuring the loudness of a signal. For example, the commonly-encountered A, B. C or Dweighted loudness measurements are spectrally weighted power measurements designed to approximate the frequency selectivity of the ear, subject to different circumstances. These measurements are attractive because they correlate much better with loudness than the raw signal power whilst retaining a simplicity which makes them easy to measure and manipulate.

Summary of the Invention

The invention in its various aspects provides a method and apparatus as defined in the appended independent claims. Preferred or advantageous features of the invention are set out in dependent sub-claims.

The invention consists of a method to equalise a signal with an equaliser that would preserve the loudness of a given nominal input signal. This allows the signal's tonality to be changed without affecting the loudness provided that the nominal input signal is reasonably representative of the actual signal. The invention can be applied to one or more channels of audio simultaneously and can be applied either to analogue or to digital audio signal formats. The signal loudness may be measured using any specified loudness measuring function.

In a preferred embodiment of the invention a conventional equaliser is used with an additional gain element such that the combined effect of the conventional equaliser and the additional gain restores the loudness of the given nominal input signal. The additional gain adapts automatically to changes in the frequency response of the conventional equaliser.

In a further preferred embodiment of the invention the additional gain is calculated by measuring the effect of the conventional equaliser on the loudness of the nominal input signal.

In an alternative preferred embodiment of the invention the additional gain is calculated by theoretical analysis of the effect of the conventional equaliser on the loudness of the nominal input signal.

In a preferred embodiment of the invention a standalone unit processes one or more audio streams provided by one or more audio inputs and outputs. The unit contains signal processing hardware that applies the method to the audio streams.

In another embodiment of the invention a computer is used to apply the method to one or more digital audio streams. In this case the audio can be stored on a hard disk or streamed via the computer's audio input/output hardware.

Description of a specific embodiment of the invention A specific embodiment of the invention will now be described by way of example with reference to the accompanying figures (1) and (2). This example shows how the additional gain can be calculated by measurement.

Figure ( 1) illustrates a three band parametric equaliser showing the controls for the low shelf, parametric and high shelf, and implemented as a standalone unit for typical sound studio use. Figure (2) shows the signal flow diagram of the loudness-preserving version of this equaliser as follows: Item 1 is the audio input; in this case this is an analogue input.

Item 2 generates the nominal input signal; in this case it is a pink noise generator.

Item 3 is a conventional equaliser that implements a three band parametric equaliser on the input signal.

Item 4 is a conventional equaliser that implements a three band parametric equaliser with an identical frequency response to item 3, but which acts on the nominal input signal.

Item 5 is the set of front panel controls of the 3 band parametric equaliser which dictate the frequency responses of Items 3 and 4.

Item 6 measures the loudness of the nominal input signal according to some function; in this case the function is the 'A'-weighted power measurement.

Item 7 measures the loudness of the nominal output signal according to the same function as used by item 6.

Item 8 calculates the additional gain required to make the normalised output signal have the same loudness as the normalised input signal, thereby restoring the loudness.

Item 9 applies this additional gain to the actual audio signal path.

Item 10 is the audio output signal.

Description of a second specific embodiment of the invention A specific embodiment of the invention will now be described by way of example with reference to the accompanying figures 3, 4 and 5. In this example the additional gain is calculated theoretically rather than by measurement. In addition it illustrates how the nominal input signal loudness can be calculated a-priori.

Figure (3) illustrates a computer that implements a finite impulse response equaliser to process audio from an external source onto its hard disk. Item 11 is the digital audio input to the computer. Item 12 is computer's audio input/output card. Item 13 is the computer's CPU and Item 14 is the computer's hard disk storage that is used to store the results for later playback.

Figure (4) illustrates a typical user interface that allows the user to create an un- normalised frequency response graphically. The frequency response curve is expressed as a frequency-dependent gain. The curve therefore shows the gain in decibels (dB) versus frequency. On this system the curve is defined by a set of gain values spaced equally in frequency between OHz (henceforth "dc") and the upper frequency limit of the system, determined by its sampling rate (henceforth "Nyquist").

Computer drawing tools are used to manipulate the curve to achieve the desired frequency response.

Figure (5) illustrates a suitable method to implement the frequency response described in Figure (4) on a computer system as shown in Figure (3) as follows Item 15 is the un-normalised frequency response as defined above. It is expressed as decibel gains at N /2 + 1 equally spaced frequency points between dc and Nyquist inclusive. At a given time t the k th point on this curve is AdB (t, k) . Item 16 converts this gain from decibels to an absolute gain using the following formula: A ( t, k) = l Owl' k)/20 Item 17 converts this frequency response A(t,k)to an impulse response a(t,k)by taking the inverse discrete Fourier transform of the frequency response. This makes the assumption that there is an implicit symmetry in the frequency response such that A (I, iN - k) = A (I, k) where i is any integer.

Item 18 is the digital audio input, x(t).

Item 19 convolves the digital audio input with the impulse response to create the un- normalised filter response y(t) such that N-l y(t) = a(t,) x(t-) r=0 Item 20 is the a-priori calculated nominal input loudness V' . Its derivation proceeds as follows: Hypothesise a pink nominal input signal. Let Fs be the sample rate and = N then, the power spectral density of our nominal input signal is given by: |x (ki)|2 = 1 We are using an 'A' weighted power measurement as the loudness measurement, so given the spectral power response of the 'A' weighted filter is H (f) the nominal input loudness can be approximated in the frequency domain by V, = H (ki)|X (kit)| k=l = H(k2) k=' kit This value can be pre-calculated provided that the nominal input signal and the loudness measuring function are fixed at design time. s

Item 21 calculates the nominal output loudness,Vy (I) . Let the power spectral density of the nominaloutputsignalbe |Y(t,k)| then |Y (t'k)| = |A (I, k)| |X (kit)| and therefore the nominal output loudness is calculated y: Vy (I) = H (ki)|Y(t,k2)| k=l = H (kit) |A (t, k)| |X (kit)| k=l N/2 H(k4)|A(t,k)| =v k=! k; Item 22 calculates the additional gain that would make the nominal output signal's loudness match the nominal input signal's loudness, correcting the loudness imbalance; g(t)= Item 23 applies this gain to the actual audio stream, so Y (t)=g(t)y(t) N-l = g(t)Ea(t,).x(t-) r=0 Item 24 is the audio output which is passed the signal y'(t) Features to note about this method are that the nominal input signal is never explicitly generated, and that the nominal input loudness can be calculated a-priori.

Claims

Claims 1. A method of equalising an audio signal such that a given

loudness measurement function of a given nominal input signal is preserved.
2. A method according to claim 1, in which the nominal signal is pink noise.
3. A method according to claim 1 or 2, characterized by the use of the unweighted power as the loudness measurement function.
4. A method according to claim 1 or 2, characterized by the use of a weighted power as the loudness measurement function.
5. A method according to claim 4 where the weighting is any of the standard 'A' B' 'C' or 'D' weighting curves.
6. A method according to claim 1, 2, 3, 4 or 5 where the audio signal is equalised by a conventional equaliser with an appropriate additional gain.
7. A method according to claim 6 where the additional gain is calculated from the nominal input loudness and the nominal output loudness, where the nominal input loudness is the loudness of the nominal input signal and the nominal output loudness is the loudness of the nominal input signal after it has been passed through a conventional equaliser with a frequency response identical to that of the conventional equaliser used on the audio signal.
8. A method according to claim 7 where the nominal input loudness is found by measuring the loudness of the nominal input signal.
9. A method according to claim 7 where the nominal input loudness is found by theoretical analysis of the nominal input signal and the loudness measurement function.
10. A method according to claim 9 where the loudness of the nominal input signal is calculated a-priori.
11. A method according to claim 7, 8, 9 or 10 where the nominal output loudness is found by measuring the loudness of the nominal input signal after it has been equalised by the conventional equaliser.
12. A method according to claim 7, 8, 9 or 10 where the nominal output loudness is found by analysis of the effect of the conventional equaliser on the nominal input signal.