Instantaneous (analogue) compression of speech signals

[John Woodgate]

I read in sci.electronics.design that John S. Dyson <[email protected]>

Of course, this is fairly far off topic WRT speech processing, so I
won't bother you with more off topic info (unless someone is interested
in my latest version of my audio AGC code -- very old versions are used
in some free and probably commercial software.) The new stuff
(developed in the last several years) is far far better than anything
else that I have played with (or developed myself.) It is still ugly,
but could be cleaned up if there would be any demand. (It is written in
sane C++, and happily uses inline asms for P4 SSE math operations, some
take advantage of the SIMD capabilities.)

It doesn't fit with my present project, but it's very interesting and I
will keep it in mind.

 

[John Woodgate]

I read in sci.electronics.design that Rich Grise <[email protected]>
wrote (in <[email protected]>) about
'Instantaneous (analogue) compression of speech signals', on Wed, 5 Jan
2005:

If a diode clipper is unsatisfactory, would a log amp do?

Not directly, because of the problem with negative-going half-cycles.

 

[John Woodgate]

I read in sci.electronics.design that Pig Bladder
ruid.net>) about 'Instantaneous (analogue) compression of speech
signals', on Wed, 5 Jan 2005:

You're trying to build a hearing aid, without admitting that you need a
hearing aid, is this it?

No, I have hearing aids and I make no secret of them. What I'm doing is
about:

- the effects of (non-linear) signal degradations on speech
intelligibility;

- the anomaly that some 'degradations' actually increase
intelligibility.

 

[budgie]

No, I have hearing aids and I make no secret of them. What I'm doing is
about:

- the effects of (non-linear) signal degradations on speech
intelligibility;

- the anomaly that some 'degradations' actually increase
intelligibility.

John, have you looked at the circuits used in commercial two-way radio
equipment? It varies from some pretty harsh diode types (with predictably high
output distortion) to some which do have the effect of improved "effectiveness"
through a better (output) average-to-peak ratio. If it will help (it's a bit
hard to actually describe the circuit) I could send you a scan of the relevant
part from one type that we found surpisingly good when overdriven by a large
margin.

 

[John Woodgate]

John, have you looked at the circuits used in commercial two-way radio
equipment?

Only as reported in general books on design.

It varies from some pretty harsh diode types (with predictably high
output distortion)

I have a fair bit of information on them, and some experience. Putting
resistance in series with the diodes is helpful, as is attenuating the
low frequencies and restoring them after clipping, but the restoration
needs care, otherwise phase-shift 'restores' the peaks (actually creates
new ones)!

to some which do have the effect of improved "effectiveness"
through a better (output) average-to-peak ratio. If it will help (it's a bit
hard to actually describe the circuit) I could send you a scan of the relevant
part from one type that we found surpisingly good when overdriven by a large
margin.

Yes, please. Email JMW[at]JMWA[dot]demon.co.uk

 

[Jim Thompson]

I read in sci.electronics.design that John S. Dyson <[email protected]>


It doesn't fit with my present project, but it's very interesting and I
will keep it in mind.

Recalling an ancient project... isn't the ideal clipping case
something like (for example)... a 2dB change in the input produces a
1dB change in the output?

...Jim Thompson

 

[Rick]

I read in sci.electronics.design that Rich Grise <[email protected]>
wrote (in <[email protected]>) about
'Instantaneous (analogue) compression of speech signals', on Wed, 5 Jan
2005:
Not directly, because of the problem with negative-going half-cycles.

What about a "true" log amp (aka Log Video)? These have a sort of
S-shaped response that's symmetrical about 0V.

 

[Nico Coesel]

I read in sci.electronics.design that Adrian Jansen <[email protected]>


Yes, and low-cost. FFT and DSP solutions are unlikely to be justifiable
on a cost/benefit basis, but a couple of diodes and a quad op-amp is a
different matter.

In that case, use a voltage controlled amplifier and control the
voltage by the filtering the output of a peak detector. I've used
these circuits many times with excellent results.

 

[Nico Coesel]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Tue, 4 Jan 2005:

Sure, but that comes later. I'm first looking for different techniques
to try.

In fact, it's easy to define the critical points in millivolts or
whatever using sine wave signals; it's another matter to make
*meaningful* measurements of the speech signals.

For that you will need special equipment and or software. One of the
simplest measurement is STI which stands for Speech Transmission
Index. This will give you a number which says how good or poor the
speech is conveyed. Bruel & Kjaer make equipment to do these sort of
measurements.

 

[Nico Coesel]

John, have you looked at the circuits used in commercial two-way radio
equipment? It varies from some pretty harsh diode types (with predictably high
output distortion) to some which do have the effect of improved "effectiveness"
through a better (output) average-to-peak ratio. If it will help (it's a bit
hard to actually describe the circuit) I could send you a scan of the relevant
part from one type that we found surpisingly good when overdriven by a large
margin.

Someone once told me that mixing a speech signal with itself but
shifted up one octave also makes it easier to understand. Never tried
it though. I think you are referring to a similar effect.

 

[John Woodgate]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Wed, 5 Jan 2005:

Recalling an ancient project... isn't the ideal clipping case something
like (for example)... a 2dB change in the input produces a 1dB change in
the output?

There was a time when anything more than that introduced unpleasant
artefacts into the signal, but that doesn't happen with modern designs.

 

[John Woodgate]

I read in sci.electronics.design that Nico Coesel <[email protected]>

In that case, use a voltage controlled amplifier and control the voltage
by the filtering the output of a peak detector. I've used these circuits
many times with excellent results.

This is not instantaneous. There is inevitably a time-constant
associated with the rectifier filter capacitor. I, too, have used this
technique, but it isn't what I want for the present project.

 

[John Woodgate]

I read in sci.electronics.design that Rick <[email protected]>

What about a "true" log amp (aka Log Video)? These have a sort of
S-shaped response that's symmetrical about 0V.

I wouldn't call that a true log amp. How is it done? My diode clippers
produce a response like that, but the result is not necessarily the best
that can be done. Possibly the theoretical best is a cube law, since
that introduces only third harmonic distortion and third-order
intermodulation.

 

[John Woodgate]

I read in sci.electronics.design that Nico Coesel <[email protected]>

For that you will need special equipment and or software. One of the
simplest measurement is STI which stands for Speech Transmission Index.
This will give you a number which says how good or poor the speech is
conveyed. Bruel & Kjaer make equipment to do these sort of measurements.

I didn't mean measurements of intelligibility, I meant measurements of
voltages. There are numerous pitfalls even in that apparently simple
measurement.

The Bruel and Kjaer RASTI box is long gone. There are numerous STI and
RASTI meters available now, but there is an increasing weight of
problems and anomalies associated with STI. The study of this is another
project in which I have a significant interest, and I am hosting a
meeting on the subject on Friday.

 

[Reg Edwards]

I assume you are referring to speech clipping in order to increase the
received volume level without exceeding the peak-power capacity of an AM or
SSB transmitter.

The following is based on personal design experience albeit 25 years back.

It really does work.

There is a microphone gain control pot. Following the microphone amplifier
is a simple 400 Hz to 3 KHz filter. The filter should be correctly
terminated with Ro to minimise over-swing on sharp speech transients.
Maximum available filter output volts being about 6 volts peak-to-peak.

There is then a 10K resistor followed by a pair of back-to-back small
signal, high-gain transistors. The transistors, such as BC109's, are diode
connected with base connected to collector. The transistors behave as
clipping diodes with a much sharper than normal knee transition. The
sharper the better!

The maximum output from the clipping circuit is plus or minus 0.6 volts. The
amplifier following the clipping circuit should have a high input impedance,
100K or greater, so as not to interfere with clipping action.

Coupling capacitors following the clipper should be large enough to pass the
lowest audio frequencies passed by the 400 Hz to 3 KHz filter without
attenuation.

Then follows the remainder of the transmitter, either AM or SSB. A second
gain control is needed after the clipping operation to set the transmitter
drive level.

At a clipping level of 6 dB on speech peaks, speech is highly intelligible
with hardly any distortion. Music is quite distinguishable. 6 dB is
equivalent to a 4-times increase in transmitter power.

Clipping levels of 10 dB or more can be successfully used to improve
received signal to noise and interference ratios.

===============================

There is a more complicated arrangement which marginally improves clipping
performance but which slightly changes the tonal quality of speech. It may
sound like a different person speaking in a different room. It requires a
balanced modulator and an additional IF side-band crystal filter at 455 KHz,
identical to the crystal filter in the main SSB transmitter.

The clipping is done at IF between the pair of crystal filters.

These excellent signal processing techniques went out fashion with
homebrewing and when citizens' band came in.

 

[Jim Thompson]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Wed, 5 Jan 2005:

There was a time when anything more than that introduced unpleasant
artefacts into the signal, but that doesn't happen with modern designs.

Keep us posted... this is becoming a fascinating subject.

...Jim Thompson

 

[Jim Thompson]

I read in sci.electronics.design that Rick <[email protected]>

I wouldn't call that a true log amp. How is it done? My diode clippers
produce a response like that, but the result is not necessarily the best
that can be done. Possibly the theoretical best is a cube law, since
that introduces only third harmonic distortion and third-order
intermodulation.

How about TANH?

...Jim Thompson

 

[Ken Smith]

I read in sci.electronics.design that Rick <[email protected]>

I wouldn't call that a true log amp. How is it done? My diode clippers
produce a response like that, but the result is not necessarily the best
that can be done. Possibly the theoretical best is a cube law, since
that introduces only third harmonic distortion and third-order
intermodulation.

How about this:


CA3080
---------!+\
! >---------+---------
--!+/ !
! !
---------------+
!
---
---
!
GND

The distortion should only be odd order harmonics since the CA3080 puts
out nearly the same current in each direction. Since it is slew rate
limiting, it is equivelent to a trebble boost, clip and then trebble cut.
This should reduce the amplitude of the harmonics. You can vary the slew
rate slope with the pin 5 current on the CA3080.

 

[Nico Coesel]

I read in sci.electronics.design that Nico Coesel <[email protected]>


This is not instantaneous. There is inevitably a time-constant
associated with the rectifier filter capacitor. I, too, have used this
technique, but it isn't what I want for the present project.

The circuit I used, uses 2 LM393 comparators (for positive and
negative peaks). The outputs are open drain and can discharge a small
timing capacitor quite fast.
If you need a faster way, you need to go digital so you can use a
feed-forward approach. A PIC processor and an 8kHz u-law serial codec
should be enough.

 

[John Woodgate]

I read in sci.electronics.design that Jim Thompson
4ax.com>) about 'Instantaneous (analogue) compression of speech
signals', on Wed, 5 Jan 2005:

How about TANH?

Sadist! Do you KNOW the series expansion of tanh? I realise that it is
easily implemented with a long-tailed pair, but I can't calculate the
harmonic spectrum; I'll have to write a Mathcad script.

But thanks for the suggestion - I think.