EchoTalk — the Apple II Echo II speech synthesizer, on your PC

This turns text into a WAV file that sounds like an Apple II with a Street Electronics Echo II card in it — the speech synthesizer a lot of blind computer users heard first, back when a talking computer was a new idea.

It does not imitate that sound. It runs the real thing: the actual Textalker software from 1981 and 1985, executing on an emulated 6502 processor, driving an emulation of the TMS5220 speech chip the card was built around. The bytes going into the speech chip are the same bytes a real Echo II would have received. Speech produced this way has been checked against a recording of real hardware and matches it exactly.

This is a sneak peek. The real goal is a screen reader voice for NVDA. This program is how you can hear it in the meantime.

Written by an AI

The code was written by Claude, Anthropic’s AI, working as an agentic coder through Claude Code. A human — Jayson Smith, @jaybird110127 on dragonscave.space — directed the work, supplied the Textalker disks, made the design decisions and, importantly, did the listening. Several real bugs in this were found by ear and would not otherwise have been caught, including a speech-pacing fault that had survived five sessions of automated measurement saying everything was fine.

Mentioned because you deserve to know how something you are running was made, not as a boast. Treat it as you would any other code from an unfamiliar source.

What is in this folder

File	What it is
`say64.exe`	The program, 64-bit
`say32.exe`	The program, 32-bit — use this on older Windows
`Textalker 3.1.3.bin` + `Textalker 3.1.3 Loader.bin`	Textalker 3.1.3 (1985)
`Textalker 1.3.bin` + `Textalker 1.3 Loader.bin`	Textalker 1.3 (1981)
`THIRD_PARTY_LICENSES.md`	Required licence notices — keep this with the program

Nothing needs installing and nothing is written outside this folder. Both programs are self-contained; there are no DLLs to copy.

Use say32.exe if you are unsure. It runs on every version of Windows. say64.exe needs Windows 10 or newer unless you have the Universal C Runtime update installed.

The two voices

The two Textalker versions sound noticeably different, and both are included because both are worth hearing.

3.1.3 is the later, more polished release, and the one most people remember.
1.3 is from 1981 and rougher around the edges. It also ignores a couple of commands the newer one understands, noted below.

They come in pairs — each version has a loader file and a main file, and the two must match. Mixing a 3.1.3 loader with the 1.3 program will not work, because the loader is what says where the program belongs in memory. say prints which version it detected when it starts, so you can always check.

Running it

say writes a WAV file. It does not play audio itself — open the resulting file in whatever you normally use.

The general shape is:

say64 [options] <loader file> <Textalker file> <text> <output.wav>

The filenames here contain spaces, so put quotes around them.

Say something

say64 "Textalker 3.1.3 Loader.bin" "Textalker 3.1.3.bin" "Hello there." hello.wav

Read a text file instead

Use --file and leave the text out:

say64 --file story.txt "Textalker 3.1.3 Loader.bin" "Textalker 3.1.3.bin" story.wav

Plain text files work best. Modern typographic characters — curly quotes, dashes, accented letters, emoji — are converted to something Textalker can pronounce, since it predates all of them and only knows the original 128 ASCII characters.

Hear the older voice

say64 "Textalker 1.3 Loader.bin" "Textalker 1.3.bin" "Hello there." old.wav

Options

Option	What it does
`--file PATH`	Read the text from a file. Leave out the text argument.
`--compressed`	Textalker’s faster speech mode. It works by dropping bits of each sound rather than playing faster, so it has a character of its own.
`--speed MULT`	Speed, 0.25 to 4. Pitch does not change at all. This is the one you probably want.
`--frame-rate N`	The speech chip’s own four fixed speeds, 0 to 3. Roughly 1x, 1.3x, 1.9x and 3.4x, pitch unchanged, but rougher. `--speed` is smoother.
`--clock MULT`	Pretends the speech chip runs at a different clock, e.g. `1.5`. Changes speed and pitch together — the sped-up-tape effect. Range 0.25 to 4.
`--pitch N`	Voice pitch, 0 to 63. Default 24.
`--volume N`	Volume, 0 to 15. Default 12.
`--word-delay N`	Pause between words, 0 to 15. Default 0.
`--repeat-filter N`	0 to 99, default 99. See below.
`--rate HZ`	Sample rate of the WAV, e.g. `22050`. Default 8000, which is what the real card produced.
`--chunk N`	Split long lines every N characters. Default 80.
`--no-chunk`	Never split. Prints a warning; see below.
`--raw`	Send bytes to Textalker untouched, skipping the conversion of modern characters.

The four speed controls, and why there are four

They do genuinely different things and can be combined:

--speed makes speech faster or slower without touching the pitch at all, smoothly, anywhere from a quarter speed to four times. This is the one you probably want. It works by changing how fast the chip walks through each sound rather than by playing anything faster, so the voice keeps its character.
--frame-rate is the speech chip’s own four fixed steps. Also pitch-preserving, but coarse, and the top step is rough.
--clock speeds it up and raises the pitch, exactly as overclocking the real chip would. Chipmunks.
--compressed is Textalker’s own idea of fast speech, which it achieves by leaving sounds out.

say64 --speed 1.8 "Textalker 3.1.3 Loader.bin" "Textalker 3.1.3.bin" "Reading quickly now." fast.wav
say64 --clock 1.6 "Textalker 3.1.3 Loader.bin" "Textalker 3.1.3.bin" "Help, I am a chipmunk." silly.wav
say64 --pitch 8 --rate 22050 "Textalker 3.1.3 Loader.bin" "Textalker 3.1.3.bin" "Low and slow." deep.wav

Higher --frame-rate values also change the texture of the voice, not just its speed, because the chip has less time to slide between sounds. 3 is fast and rough; try 1 or 2 first.

The repeat filter

Textalker has a feature that collapses runs of the same character, so a line of ***** is not read out as “star star star star star”. It cannot tell decoration from content, though, so at its original setting it also turns EEEEEEEEE into EE.

It is switched off by default here. --repeat-filter 2 or similar puts the original behaviour back if you want to hear it.

Echo/Textalker command reference

If you’re preparing a file to send to the Echo and want to embed any of these commands, you’ll need to use a hex editor or some other editor which lets you insert control characters.

All prefixed with Ctrl-E (\x05) unless noted: nP pitch 0-63 (default 24), nF flatness, nV volume 0-15 (default 12), nD inter-word delay 0-15 (v3.1.3 only), C/E compressed/expanded, T talk-only, nR repeat filter, \x16 (Ctrl-V) phoneme mode terminated by CR.

Things worth knowing

A single character on its own is announced properly. Type just a comma and you will hear “comma”. Textalker would normally say nothing.
--word-delay does nothing on Textalker 1.3. That version never implemented the command and quietly ignores it. Not a bug here.
Long text is split at sensible places. Sentences and phrases are broken at punctuation or between words, never mid-word, because Textalker’s own line buffer would otherwise cut wherever it happened to fill up.
The pauses are real. Textalker takes a moment to think before each phrase, and on a real Apple II you would hear that. Most of it is removed here, but the natural pauses after commas and full stops are deliberately kept.

What changed since the first release

If you have an earlier copy of say, these are the differences worth knowing about:

A single character no longer says “return” after it. In the first release, asking for a lone comma got you “comma return”. That was a real bug and it is fixed.
--speed is new, and is a better speed control than anything that was there before: smooth, and it genuinely does not move the pitch.
Long word delays and slow speeds used to mangle the speech. The emulator gave up part-way through a character and said nothing about it. Fixed, and it now reports itself if it ever happens again.
Ctrl-D driver commands, below, are new.

Driver commands, with Ctrl-D

Ctrl-E commands go to Textalker. There is a second set, introduced by Ctrl-D (), that controls the emulator itself. Same shape: an optional number, then a letter.

Command	What it does
`1.5S`	Speed, as `--speed`
`0.75C`	Chip clock, as `--clock`
`2F`	Frame rate, as `--frame-rate`
`0B`	Turn chunking off; `80B` puts it back
`1R`	Raw text on; `0R` off
`F`	A bare letter restores that setting’s default
	One literal Ctrl-D character, spoken rather than obeyed

Each command ends the current phrase, so what comes before it is spoken at the old setting and what follows at the new. Note that the letters mean different things from the Ctrl-E ones: Ctrl-E F is monotone, Ctrl-D F is frame rate.

Licences

THIRD_PARTY_LICENSES.md covers the emulation code, which is public domain and BSD-licensed, and must stay with the program.

The Textalker files themselves are the original commercial software from Street Electronics, with the 3.1.3 release also carrying an American Printing House for the Blind copyright. They are not covered by those licences — say runs the genuine article rather than a reimplementation, and those files remain the property of their copyright holders.