Re: No speech in the alpha Installer, and speech freeze in Bookworm

On Sun, Jun 29, 2025 at 05:23:49PM +0200, Cyril Brulebois wrote:
>Hi,
>
>Roland Clobus <[email protected]> (2025-05-27):
>> I agree, but we would support a few more sound cards now.
>> 
>> For the cAVS sound card firmware-intel-sound might be required, and no
>> additional kernel modules. The firmware-intel-sound.patterns file in the
>> netinst image does not contain that identifier (even after the update),
>> however the patterns file in the live images does contain that identifier
>> [1].
>> Interestingly enough, the identifier 8086:51ca (Intel cAVS) is mentioned in
>> both patterns files on the live images. I read some kernel code, and
>> determined at firmware-intel-sound would be required, but I forgot to write
>> down exactly how I came to that conclusion.
>> 
>> I ended up with [3]:
>> for _ALIAS in $(dpkg-query -L ${_LINUX_KERNEL_PACKAGE} | egrep
>> 'kernel/sound/soc/intel/.*\.ko(.xz)?$'); do /usr/sbin/modinfo ${_ALIAS} |
>> awk '/^alias:/ { print $2 }'; done | sort -u | awk '{ gsub(/[*]/, ".*", $1);
>> printf "^%s$\n", $1 }' > ${_FW_DEP11_DIR}/${_PACKAGE}.patterns
>> 
>> which parses different kernel modules compared to the logic for
>> firmware-sof-signed:
>> for _ALIAS in $(dpkg-query -L ${_LINUX_KERNEL_PACKAGE} | egrep
>> '(kernel/sound/soc/sof/.*\.ko(.xz)?|/snd-soc-.*sof.*\.ko(.xz)?)$'); do
>> /usr/sbin/modinfo ${_ALIAS} | awk '/^alias:/ { print $2 }'; done | sort -u |
>> awk '{ gsub(/[*]/, ".*", $1); printf "^%s$\n", $1 }' >
>> ${_FW_DEP11_DIR}/${_PACKAGE}.patterns
>> 						
>> On top of this, there are 2 scenarios to consider:
>> 1) Have espeakup work during the installation
>> 2) Ensure that the installed system has the correct firmware files
>> 
>> For 1) the additional firmware is required in the initrd
>> For 2) the .patterns files are required (the firmware-intel-sound.patterns
>> file was not generated
>
>I thought I had followed up to this but it seems that didn't happen…
>
>At the moment, I'm very much not convinced that the initial debian-cd
>patches (besides fixing firmware inclusion logic) are correct, so it would
>make sense to me to get the firmware-intel-sound addition reverted, so
>that the next release's focus is on trying to ensure i915 gets loaded
>successfully (hopefully the existing firmware-sof-signed integration gets
>improved, with speech synthesis starting right off the bat).

OK. I took the patch here for debian-cd on the assumption that you
thought it was correct/helpful. Happy to revert if you'd prefer that.

>In a follow-up release (possibly after 13.0 depending on the release
>team's timing when it comes to the full freeze and the initial trixie
>release), we can try and come up with proper firmware-intel-sound support,
>which would mean (as you identified) making sure speech synthesis works
>within d-i, plus ensuring the relevant package(s) get deployed in the
>installed system:
> - Having one without the other would only mean wasting people's time, I
>   fear.
> - Plus the huge extra cost for the initrd (size-wise) is clearly not
>   worth it at this time.

ACK.

>Also, I suggest turning the firmware-intel-sound topic into a dedicated
>bug report, instead of keeping it entangled with firmware-sof-signed in
>this overlong thread plus merge requets / commit comments on Salsa…

*Definitely*