Open Bug 1767254 Opened 2 years ago Updated 1 year ago

web speech - wikipedia - ff does not emits the speech of accented wovels

Categories

(Core :: Web Speech, defect)

Firefox 99
x86_64
Linux
defect

Tracking

()

People

(Reporter: golonlutoj, Unassigned)

Details

Attachments

(1 file)

Steps to reproduce:

System: Linux UBUNTU 20.04 LTS
I am using speech synthesis in the read panel of FF.
I set the linux configuration of the speech dispatcher as "espeak-mbrola-generic.conf".
The Wikipedia page that is not working is (ITALIAN ) https://it.wikipedia.org/wiki/Contrabbasso

Actual results:

Speech synth is working almost correctly, but when there is an accented wovel, the speech of this wovel is not emitted, and the word it belongs results incomplete.
I made a test, tracing the text that FF feeds to the espeak software, and I found that the wovels are subsistuted with a '?' letter.

Expected results:

Accented wovels have to be emitted by synth.
eSpeak software works regularly if it is used both with a text at command line, or text in a file: all accented wovels are emitted.

Component: Untriaged → Reader Mode
OS: Unspecified → Windows 10
Product: Firefox → Toolkit
Hardware: Unspecified → x86_64

reader mode just uses the web speech component in gecko, so this will affect anything using web speech. It probably has to do with the Linux webspeech implementation.

Component: Reader Mode → Web Speech
OS: Windows 10 → Linux
Product: Toolkit → Core

I would like to confirm this bug if it's possible to reproduce it on my Ubuntu 22 machine, but I do not understand how I can make it read from a firefox webpage. Can you explain how to make it do that?

Thanks!

Flags: needinfo?(golonlutoj)

(In reply to Bodea Daniel [:danibodea] from comment #2)

I would like to confirm this bug if it's possible to reproduce it on my Ubuntu 22 machine, but I do not understand how I can make it read from a firefox webpage. Can you explain how to make it do that?

Thanks!

Enter reader mode, then click the headphones ("listen") icon, then click the "play" icon.

Flags: needinfo?(golonlutoj) → needinfo?(daniel.bodea)

I was not able to reproduce this issue.

I was able to discover that the speech-dispatcher should be set up as OS-level by using the spd-config command in the terminal.

Attached is the configuration made.

Considering I can't reproduce it, I imagine I'm setting it up differently than you are.
Does the issue still reproduce in your case?
How (exactly) have you set it up?
Which version reproduces it?
Does it happen in safe mode (info here)?

Flags: needinfo?(daniel.bodea) → needinfo?(golonlutoj)

Some additional test.

  1. updated my xubuntu --> updated to FF101 : same bad results

  2. used the troubleshoot mode : same bad results

  3. info: my general language setup for my xubuntu PC is in Italian language

  4. for the synth, I use the voice "it4" (female); to have this voice in FF, I edited the file /etc/speech-dispatcher/modules/espeak-mbrola-generic.conf
    and added the following lines
    AddVoice "it" "MALE1" "it3"
    AddVoice "it" "FEMALE1" "it4"

  5. I didnt use the spd-config tool, I just manually edited the config files. I found instructions in stackoverflow and similar sites

  6. in the file /etc/speech-dispatcher/speechd.conf I leaved uncommented only the following line
    AddModule "espeak-mbrola-generic" "sd_generic" "espeak-mbrola-generic.conf"

  7. in the file /etc/speech-dispatcher/modules/espeak-mbrola-generic.conf I tested with the following line
    GenericExecuteSynth
    "echo '$DATA' | espeak -v mb-$VOICE -s $RATE -p $PITCH $PUNCT -q --stdin --pho | mbrola -v $VOLUME -e /usr/share/mbrola/$VOICE/$VOICE - -.au | $PLAY_COMMAND"
    this was the original config line

then tested the following (simplified)
GenericExecuteSynth "echo '$DATA' | espeak -v mb-$VOICE -s $RATE -p $PITCH $PUNCT --stdin"

and finally I wanted to see what FF is feeding to the synth process, so I just echoed the output text from FF to a file
GenericExecuteSynth "echo '$DATA' >> /home/gp/ff.txt"

The resulting ff.txt file revealed that all accented wovels are all substituted with a ? (question mark) character, and this explains why the synth child process skips them in the audio output.

  1. I loaded in FF a plain text file containing a text with accented wovels (url is some like file:///home/gp/mytestfile.txt ... ). Reader mode is available also for this type of files.
    Then I tested the synth, but again bad results. I think this means that it is not an HTML-related problem.
Flags: needinfo?(golonlutoj)
Severity: -- → S3

These configurations are far too complicated to attempt reproduction manually, but hopefully, now it was enough information for the development team to address it. The reporter has reconfirmed the reproduction in later versions than the one originally provided and on a different version of Ubuntu. I will confirm this issue based on the reporter's response. Please NeedInfo me if I can provide test results on Ubuntu 21/22 on very specific reproduction steps.

Thank you for your contribution!

Status: UNCONFIRMED → NEW
Ever confirmed: true
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: