When the Windows Subsystem for Linux first came online, I installed it immediately. Using a native Windows screen reader on a Linux system is not ideal. Installing a Linux screen reader was out of the question—there was no audio support.

I wanted to revisit the state of play—*playing audio*, that is—on this quiet Sunday. I had nothing planned, and my ride to church fell through due to illness.

I had many questions. I turned to Copilot on my Windows machine for help. I also received encouragement.

I wanted to use Emacs—it’s what I use on my Raspberry Pi. Together, Copilot and I explored a couple of options:

• Using Orca with GNOME in WSL
• Installing Emacspeak to run on Windows itself

I hadn’t realized someone had worked out how to get Emacs speaking on Windows. You can read about it here: Emacspeak on Windows. Unfortunately, the preset configuration used older versions of both Emacspeak and Emacs. I decided to install and configure everything myself.

I found myself confronted with a directory of files to download. They were very large and took a long time. When I unzipped the archive, there were over 60 tarred files. I expected an executable, but didn’t find one at first. After reviewing the online directory, I discovered the `.exe` file about 12 lines down. It had been updated in 2021, but its position in the list hadn’t changed. This was still an older version of Emacs, but I thought I could live with it. I downloaded the executable and installed it.

I’m spoiled—my Raspberry Pi runs Arch Linux, so I’m used to the latest and greatest. Arch is a rolling release, so packages are constantly updated.

Next, I downloaded Emacspeak. I extracted all the files and placed them in the Emacs folder. I was puzzled. The next thing I always do with Emacspeak on Linux is:

make config  
make  

before setting up the speech server.

Copilot’s instructions didn’t mention this, so I went onto the server itself.

I started reading how to install the server. It was complicated—downloading and configuring many more packages. I didn’t understand much of the technical detail, and it looked like the project had been abandoned with Emacspeak 43. I’m running version 60.

Time to pull myself out of this rabbit hole.

When I cleaned out all the files, there were over 13,000 of them.

My last hope for a talking Emacs installation on Windows was WSLg (also known as WSL2). I installed Emacs, cloned Emacspeak, and built it:

sudo apt update -qq  
sudo apt install emacs  
git clone https://github.com/tvraman/emacspeak.git  
make config  
make  

Now it was time to set up the speech server. My only real choice was espeak, an old Linux friend:

sudo apt install espeak tcl tclx  
cd ~/emacspeak/servers  
espeak  

Type `C-d` to exit the cpp prompt.

Espeak worked as expected:

espeak "Hello there."  

Espeak echoed “Hello there.” (Without the quotation marks, of course.)

So far, so good.

I ran Emacs and told it to load Emacspeak:

emacs -l ~/emacspeak/lisp/emacspeak-setup.el  

Initially, Emacspeak spoke—but then went silent. It didn’t matter whether my Windows screen reader was running or not. The audio in Emacs just fell apart.

On my Chromebook’s Linux container, I use speechd.el. I copied my `.emacs` file from the Chromebook and installed the Ubuntu package:

sudo apt install speechd-el speech-dispatcher espeak-ng  

The nice thing about speech-dispatcher is that I don’t have to configure it. It uses PulseAudio. Espeak-ng has a slightly less robotic sound. I’ve been using speech synthesis for a very long time, and robotic speech doesn’t bother me.

Before I unregistered Ubuntu from WSL, I ran spd-say:

spd-say "Hello there."  

Although speech-dispatcher was running, spd-say was silent. Speech wasn’t going to happen.

So I’m back here on my Raspberry Pi, where speech just works—and I can write this post.

It seems strange to me that audio, which is so vital to the blind community, is so unimportant to those who develop software. Speakup is such an elegant solution for console speech and would be a great help in WSL and the Linux container on Chromebook. I use it here on the Raspberry Pi. It’s part of the Stormux image, though its default console screen reader is Fenrir. I’m just not very familiar with Fenrir, so I’ve switched to Speakup because it’s familiar and I know how to use it.

When I exit Emacs and turn Speakup back on, it says, “Hey, that’s better.” That sums up my quiet Sunday.

Here is what Copilot has to say.

Your post is a quiet masterclass in what accessibility experimentation actually feels like: hopeful, tedious, occasionally triumphant, and often absurd. Here’s what stood out to me:

When I asked Copilot if our conversations about speech and accessibility would be folded back into its general knowledge, it was quick to assure me that our conversations are private and not shared. I think this is a lost opportunity for the knowledge base for accessibility to grow. The developers of wsl, for example, will be unlikely to make audio reliability a priority.