DECtalk builds: compiling the 1990s speech synthesizer on Linux, Windows and macOS

What dectalk/dectalk actually is, and who needs it

DECtalk is a formant speech synthesizer from the 1990s and 2000s. The repository holds the source code that its original developer, Edward Bruckert, shared on the DECtalk mailing list in October and November 2015, plus code another developer added in September 2022. The project's stated goal is narrow: modern builds for that application. It is not a rewrite, not a new synthesizer, and not a service.

The audience follows from that. Someone preserving a piece of speech technology history. Someone who wants DECtalk voices on a machine that no longer runs the original binaries. Someone porting the engine to a platform the original vendors never shipped, which the develop branch already does for i386, x86_64, aarch64, Apple Silicon and Intel Macs, and iOS according to the README. The README points people who just want to hear it at a WebAssembly build in the browser and at GitHub Actions artifacts, which is the right first stop for anyone who does not intend to compile anything.

Two branches, one source dump and one working build tree

The branch layout is the most useful thing to understand before cloning. The README describes master as "literally just the source code dumped into the /src folder" and develop as the branch with working Linux and Windows binaries and MacOS/iOS support. develop is the default branch, so a plain clone gets the buildable tree, but anyone who has read older DECtalk material may be looking at master and wondering why nothing compiles cleanly.

The build system is autotools on Unix-like systems, with Visual Studio solutions for Windows and a batch-file pipeline for Visual Studio 6. The repository root also carries a Dockerfile and a docker-compose.yml, so the whole toolchain can be containerized. The Dockerfile is a two-stage build: a debian:bullseye stage installs build-essential, libasound2-dev, libpulse-dev, libgtk2.0-dev, unzip, autoconf, automake and git, copies the tree to /dectalk, then runs autogen.sh, configure and make -j in /dectalk/src. The second stage is scratch, copying only /dectalk/dist into ./dist. That is a clean separation: the compiler image never ships, and the output is just the built artifacts.

Building DECtalk on Linux, Windows and macOS

On Ubuntu the README lists the dependencies explicitly, with a note that libgtk2.0-dev is only needed for the graphical frontend, libasound2-dev for ALSA output and libpulse-dev for PulseAudio output. Drop the ones you do not need if you are building a headless synthesizer.

apt-get install build-essential libasound2-dev libpulse-dev libgtk2.0-dev unzip git

The compile itself runs from the src directory. autogen.sh generates the configure files, configure executes them, and make builds. The README notes that -j uses all cores and can be removed for a single-core build. The output lands in a dist folder, which in a fresh checkout does not exist yet.

cd src
./autogen.sh
./configure
make -j

If you would rather not install the toolchain at all, the repository ships a Compose file. Running it builds the image and copies the artifacts into a dist directory in the project root, which is the same output path as a local build.

docker-compose up

The README also documents a Docker-only path, `docker build --output=. .`, and warns that permission errors can be worked around by creating a writable output directory first, for example `mkdir output`, `chmod 777 ./output`, then `docker build --output=./output .`.

Windows has two routes. Visual Studio 2022 with the "Desktop development with C++" workload opens src/DECtalk.sln directly; the README mentions ARM64 build tools as optional and untested. Visual Studio 6 is the legacy route: add the VC98 Bin directory to PATH, then run devops/vs6/dt_buildall.bat from the project root and collate with devops/vs6/dt_copyfiles.bat. On macOS, install the Xcode command line tools, then run autoreconf -si, configure and make from src, again producing dist.

Where the documentation stops

The README is a build guide and nothing else. It does not document the command-line interface, the phoneme set, the voice selection, or how to drive the synthesizer from another program. There is no rollback or uninstall procedure, no statement of which configure flags exist beyond the GTK/ALSA/PulseAudio dependency hints, and no troubleshooting section for build failures. If you are coming to DECtalk from its reputation rather than its manual, the repository will not teach you how to make it talk.

The licence is the sharper problem. The repository root contains a file named LICENCE, but the project's licence is reported as NOASSERTION, meaning no recognized licence identifier was detected. The README does not discuss terms of use at all, and the source arrived through a mailing-list share and a preserved archive rather than through a formal release. Anyone planning to redistribute binaries or embed the engine in a product has to read that file themselves and decide whether the terms cover their use. Nothing in the repository answers the question for them.

Maintenance is another thing to weigh rather than assume. The last push to the repository was on 2026-07-21, and the most recent tagged release is from 2023-10-30. The README directs users to Discord for community support, which is where questions about build breakage would go.

DECtalk versus modern neural text to speech

The obvious alternative is a current open-source neural synthesizer such as Piper or Coqui TTS, and the difference is not quality, it is category. A neural model is trained on recorded speech and produces output that sounds like a recording of a person; it ships with model files that are hundreds of megabytes, needs a runtime like ONNX Runtime or PyTorch, and its voice is whatever the training data contained. DECtalk is a rule-based formant synthesizer: it generates speech from parameters, which is why it is small, why it runs on hardware from the 1990s, and why its output is instantly recognizable as DECtalk rather than as a person.

That distinction decides the use case. If you want a natural-sounding voice for an application, DECtalk is the wrong tool and a neural model is the right one. If you want the specific DECtalk voices, if you are studying how formant synthesis was implemented, or if you are porting the engine to a platform where a neural runtime will not fit, then the alternatives do not substitute for it. The repository's own README points browser visitors at a WebAssembly build, which is the sensible middle path for anyone who wants the sound without the build.

Who should compile it, and what to check first

Compile this if you need a local DECtalk binary on Linux, Windows or macOS, if you are porting the engine to a platform the develop branch does not cover, or if you are working on the source itself. The Docker path is the lowest-friction start because it pins the toolchain to debian:bullseye and produces the same dist directory a local build does.

Skip it if you need a supported speech API with documentation, a licence you can read at a glance, or a maintainer who will answer a bug report within a release cycle. The README does not document the runtime interface, the repository's licence is unclassified, and the last push was on 2026-07-21 with the most recent release dating to 2023-10-30.

Before you commit, do three things. Run the Compose build once and confirm that dist appears with the binaries you expect, since the README does not enumerate what gets built. Read the LICENCE file in the repository root and decide for yourself whether your distribution plan is covered. And check whether the default configure invocation builds the GTK frontend or only the engine, because the README treats libgtk2.0-dev as optional without saying what changes when it is absent.

Editorial conclusion

Adopt this repository if you need a local DECtalk binary, are porting the synthesizer to a new platform, or want the source preserved in a buildable state; the develop branch is where the Linux, Windows, macOS and iOS build files live, and the last push was on 2026-07-21. Do not adopt it if you need a supported, documented speech API with a maintainer behind it, or if you expect the README to explain the licence: it does not. Before committing, run the Docker build once to confirm the toolchain works on your machine, read LICENCE in the repository root, and check whether the frontend you want (GTK or command line) is built by the default configure invocation, because the README only mentions the GTK dependency in passing.

Frequently asked questions

Official sources

1. dectalk/dectalk on GitHub
2. Issues
3. Project website
4. README
5. Releases