[SoX-devel] 64bit audio read and write

[Matt Flax]

Hi there,

I currently use libsox to load and save audio data. I was wanting to
write and read 64bit data, however the generic sox sample type is 32 bit
(from sox.h) :
typedef sox_int32_t sox_sample_t;

The basic format of the read (and write) functions are :
size_t sox_read(
sox_format_t * ft,
sox_sample_t *buf,
size_t len
);

Can anyone tell me the basic methodology for reading/writing 64bit audio ?

thanks
Matt

[Eric Wong]

Post by Matt Flax
Hi there,
I currently use libsox to load and save audio data. I was wanting to
write and read 64bit data, however the generic sox sample type is 32 bit

Correct, sox does some calculations internally in 64-bit double,
but data which flows in between the effects is all 32-bit int.

Post by Matt Flax
typedef sox_int32_t sox_sample_t;
size_t sox_read(
sox_format_t * ft,
sox_sample_t *buf,
size_t len
);
Can anyone tell me the basic methodology for reading/writing 64bit audio ?

It would probably reworking a lot of sox internals and would
require increased memory bandwidth, hurting performance for
common <= 32-bit processing.

But, where and how are you working with 64-bit audio
and what 64-bit audio formats are there?

I'm not up-to-date with the latest technology, but even 32-bit
is far beyond the range of human hearing; and last I checked;
available ADC/DACs can't even make full use of 24-bit (nor
our ears).


On the other hand, moving to 32-bit float would make sense for
compatibility with 3rd-party plugins (LADSPA for sure, maybe
LV2?). Right now, we're constantly converting between
float/integer in effects chains which hurts performance.

[Matt Flax]

Post by Eric Wong

Post by Matt Flax
Hi there,
I currently use libsox to load and save audio data. I was wanting to
write and read 64bit data, however the generic sox sample type is 32 bit

Correct, sox does some calculations internally in 64-bit double,
but data which flows in between the effects is all 32-bit int.

Post by Matt Flax
typedef sox_int32_t sox_sample_t;
size_t sox_read(
sox_format_t * ft,
sox_sample_t *buf,
size_t len
);
Can anyone tell me the basic methodology for reading/writing 64bit audio ?

It would probably reworking a lot of sox internals and would
require increased memory bandwidth, hurting performance for
common <= 32-bit processing.
But, where and how are you working with 64-bit audio
and what 64-bit audio formats are there?

I think there are 64 bit file formats, aiff-c perhaps ? Also matlab file
formats should be supporting 64 bits.
Personally, working with filters and various signal processing
algorithms like that. It is handy being able to save double length
words, possibly which are generated and synthesised in one way or other ...

Post by Eric Wong
I'm not up-to-date with the latest technology, but even 32-bit
is far beyond the range of human hearing; and last I checked;
available ADC/DACs can't even make full use of 24-bit (nor
our ears).

120 dB SPL (threshold of discomfort) is around 20 billion times louder
then 20 dB SPL (a whisper). That is approximately 24 bits.
However, we can hear down to 10 dB SPL, or lower for young people ...
that is approximately 40 bits of dynamic range.

Actually, I am not proposing we change the bit depth of sox in its
signal processing system, however perhaps we can introduce functions to
read and write 64 bit audio ?
Something like functions to read64 and write64 perhaps ?

Post by Eric Wong
On the other hand, moving to 32-bit float would make sense for
compatibility with 3rd-party plugins (LADSPA for sure, maybe
LV2?). Right now, we're constantly converting between
float/integer in effects chains which hurts performance.
------------------------------------------------------------------------------
Attend Shape: An AT&T Tech Expo July 15-16. Meet us at AT&T Park in San
Francisco, CA to explore cutting-edge tech and listen to tech luminaries
present their vision of the future. This family event has something for
everyone, including kids. Get more information and register today.
http://sdm.link/attshape
_______________________________________________
SoX-devel mailing list
https://lists.sourceforge.net/lists/listinfo/sox-devel

[Måns Rullgård]

Post by Matt Flax

Post by Eric Wong
I'm not up-to-date with the latest technology, but even 32-bit
is far beyond the range of human hearing; and last I checked;
available ADC/DACs can't even make full use of 24-bit (nor
our ears).

120 dB SPL (threshold of discomfort) is around 20 billion times louder
then 20 dB SPL (a whisper). That is approximately 24 bits.
However, we can hear down to 10 dB SPL, or lower for young people ...
that is approximately 40 bits of dynamic range.

120 dB is outright painful and damages the ears. 100 dB is already
unpleasant.

You should also check your calculations. With your limits, the required
dynamic range would be 110 dB, which comes out to 20 bits with a bit of
margin.

[Matt Flax]

Post by Måns Rullgård

Post by Matt Flax

Post by Eric Wong
I'm not up-to-date with the latest technology, but even 32-bit
is far beyond the range of human hearing; and last I checked;
available ADC/DACs can't even make full use of 24-bit (nor
our ears).

120 dB SPL (threshold of discomfort) is around 20 billion times louder
then 20 dB SPL (a whisper). That is approximately 24 bits.
However, we can hear down to 10 dB SPL, or lower for young people ...
that is approximately 40 bits of dynamic range.

120 dB is outright painful and damages the ears. 100 dB is already
unpleasant.
You should also check your calculations. With your limits, the required
dynamic range would be 110 dB, which comes out to 20 bits with a bit of
margin.

Loading Image...
Anything above 75 dB SPL has potential to damage your hearing,
depending on exposure time. The fact that we can hear with clarity over
such large dynamic range is extraordinary isn't it !

I think the point I am making is that you can't capture 64 bit words of
data in audio files with only 32 bit word reads/writes from file. I
wasn't aware that sox didn't have any methods for 64 bit audio data file
read and writes. From a pure data perspective, no qualms right ?

Matt

[Måns Rullgård]

Post by Matt Flax

Post by Måns Rullgård

Post by Matt Flax

Post by Eric Wong
I'm not up-to-date with the latest technology, but even 32-bit
is far beyond the range of human hearing; and last I checked;
available ADC/DACs can't even make full use of 24-bit (nor
our ears).

120 dB SPL (threshold of discomfort) is around 20 billion times louder
then 20 dB SPL (a whisper). That is approximately 24 bits.
However, we can hear down to 10 dB SPL, or lower for young people ...
that is approximately 40 bits of dynamic range.

120 dB is outright painful and damages the ears. 100 dB is already
unpleasant.
You should also check your calculations. With your limits, the required
dynamic range would be 110 dB, which comes out to 20 bits with a bit of
margin.

http://www.nal.gov.au/images/diagram-of-noise.jpg
Anything above 75 dB SPL has potential to damage your hearing,
depending on exposure time. The fact that we can hear with clarity
over such large dynamic range is extraordinary isn't it !

I tend to think it's a natural result of evolution in an environment
where those sound levels occur regularly.

Post by Matt Flax
I think the point I am making is that you can't capture 64 bit words
of data in audio files with only 32 bit word reads/writes from file. I
wasn't aware that sox didn't have any methods for 64 bit audio data
file read and writes. From a pure data perspective, no qualms right ?

Adding support for some file format with 64-bit samples is easy if you
accept truncating the values to 32 bits on read and padding on write.

[Jan Stary]

Post by Matt Flax
I currently use libsox to load and save audio data. I was wanting to
write and read 64bit data, however the generic sox sample type is 32 bit
typedef sox_int32_t sox_sample_t;

If all you need to do is read and write samples,
you might be better off with libsndfile.
http://www.mega-nerd.com/libsndfile/api.html


------------------------------------------------------------------------------