Subject: Re: [ecasound] 24/96 upsampling problem
From: Kamil Wencel (wencel_AT_radion.org)
Date: Sat Nov 29 2003 - 23:35:54 EET
Kai, thanks a lot.
As soon as I have tried that with the latest libsamplerate and HQ
I will give feedback if this will be suitable for audiophile`s or not ;)
Kamil
Referring to Kai Vehmanen :
> On Wed, 5 Nov 2003, Kamil Wencel wrote:
>
>> I am having trouble upsampling a 16 bit 44.1 kHz wave file to
>> 24bit 96kHz. Sorry, if my that question may seem lame, but
>> I have no idea what I´m doing wrong.
>
> This is a good question. Earlier Ecasound releases supported
> implicit resampling (i.e. resampling when needed), but this approach had
> its problems, so with current Ecasound releases you have to
> explicitly ask for resampling. Here's an example:
>
> ecasound -f:16,2,96000 -i resample,44100,foo.wav -f:24,2,96000 -o bar.wav
>
> The syntax is somewhat verbose as Ecasound requires all audio
> input and output objects (in one chainsetup) to have the same
> samplerate. To do resampling, we use the special 'resample' input
> type to resample foo.wav from 44100 to 96000. Additionally we specifiy
> the sample format (16bit for foo.wav and 24bit for bar.wav). Both
> files are stereo/2ch.
>
> For good quality, I recommend installing Erik de Castro
> Lopo's libsamplerate [1] package. To optimize conversion quality,
> use 'resample-hq' instead of 'resample' (although this consumes
> a _lot_ of cpu power). You have to re-configure and compile Ecasound after
> installing libsamplerate.
>
> If you notice any audible errors in the conversion, makek sure you have
> the latest libsamplerate (0.0.15 or newer). Also, there are few minor bugs
> in Ecasound's side. I'm currently tracing these, and will add a few
> bugfixes to the upcoming Ecasound-2.3.2.
>
>> Once I´ve upsampled the file, its content becomes *instable*
>> especially in more dynamic parts of music. Then you hear a lot
>> of scratching and fine clicks.
>
> Resampling is tricky business. Good conversions takes a lot of CPU and
> requires complex algorithms, while a simpler converted can be very light,
> but quality suboptimal. Also, error in converter code can show up with
> only certain src-rate<->dst-rate pairs, and this further complicates
> testing.
>
> [1] http://www.mega-nerd.com/SRC/
>
> --
> http://www.eca.cx
> Audio software for Linux!
>
>
RADION
Digital Research & Innovation
Kamil Wencel
Swakopmunder Str. 1
81827 München
voice 3.1kHz : + 49 89 43746158
fax-machine : + 49 89 43746159
email : wencel_AT_radion.org
browser : www.radion.org
This archive was generated by hypermail 2b28 : Sat Nov 29 2003 - 23:34:37 EET