TL;DR: The voice coding landscape turned over almost completely between 2021 and 2024. VoiceCode, GitHub Copilot Voice, Mozilla DeepSpeech, Coqui, Dragon for Mac, and Windows Speech Recognition are all dead or deprecated. Serenade is stalled. What’s thriving: Talon with Cursorless for full hands-free control, the Whisper family for local transcription, and push-to-talk dictation tools for driving AI coding agents. This page is the current map, with an honest graveyard and what to use instead of each dead tool.
If you got here from an article about coding by voice written before 2023, there’s a good chance half the tools it recommended no longer exist. This niche has an unusually high body count, and old resource pages don’t get updated. So this is the page I wished existed: every notable voice coding and speech-to-text tool, its actual status as of August 2026, and what replaced the dead ones.
If you maintain a resource list that links to any of the tools in the graveyard below, feel free to link here instead. I keep this page current.
Why so many voice tools died at once
One shift explains almost every grave on this page.
From roughly 2015 to 2021, voice coding meant command grammars. The thing listening was an editor, so you spoke syntax: “camel case user name,” “slap,” “dedent.” Tools like VoiceCode, Silvius, Caster, and Serenade were essentially compilers from spoken commands to keystrokes, and the community driving them was largely developers with RSI who had no other option. The learning curve was weeks. The payoff was real but the audience was small.
Then two things happened in quick succession. OpenAI released Whisper in late 2022, which made accurate speech-to-text free, local, and easy to embed, instantly obsoleting a decade of fragile open-source recognition engines. And AI coding agents arrived in 2023, which changed what you say. You don’t dictate syntax to Claude Code or Cursor. You describe intent, in plain English, and the agent writes the syntax. I wrote more about that shift in the Talon comparison, but the short version: the grammar tools were solving a problem that partially dissolved, and the tools that bet on plain dictation plus AI context are the ones still shipping.
Still alive in 2026
| Tool | What it is | Platform | Status |
|---|---|---|---|
| Talon + Cursorless | Full hands-free computer control and code editing | Win/Mac/Linux | Active |
| Invoke | Local push-to-talk dictation for AI coding tools | Windows | Active (mine) |
| Wispr Flow | Cloud dictation | Mac/Win/iOS | Active |
| SuperWhisper | Local dictation | Mac only | Active |
| MacWhisper | Local transcription | Mac only | Active |
| Willow Voice | Cloud dictation | Mac/Win | Active |
| Dragon Professional | The classic desktop dictation suite, now v16 | Windows only | Active |
| Voice Access | OS-level voice control built into Windows 11 | Windows | Active |
| VS Code Speech | Local push-to-talk input for Copilot Chat | VS Code | Active |
| whisper.cpp / faster-whisper | Local STT engines | All | Active |
| Vosk | Lightweight offline STT engine | All | Active |
| nerd-dictation | Scriptable hands-free dictation | Linux | Active |
A few notes on the living. Talon plus Cursorless is the deep end: full keyboard-free control with a real learning curve, still actively developed, still the answer if RSI means your hands are off the table entirely. The dictation apps (Wispr Flow, SuperWhisper, Willow, Invoke) are the shallow end that most developers actually want in 2026, hold a key, talk, text appears. I compared them properly in the Windows dictation roundup and the Whisper desktop app guide. And Dragon deserves its asterisk: alive and genuinely good on Windows, but it’s a $699 medical-and-legal product now, and its Mac story is in the graveyard below.
The graveyard
Each entry: what it was, what happened, and what to use instead. Dates included so you can verify.
VoiceCode.io — dead
What it was: The most complete out-of-the-box voice coding system of the grammar era. Mac only, $300, built on Dragon, beloved in the RSI community.
What happened: Development and support quietly stopped years ago. The website still exists, which is worse than a 404, because people still find it and buy nothing that works on a current macOS.
Use instead: Talon. It absorbed most of the VoiceCode community and it’s the only grammar-era tool still under active development.
GitHub Copilot Voice (“Hey GitHub”) — dead
What it was: GitHub’s own voice interface experiment for Copilot, launched as the “Hey GitHub” technical preview in 2022.
What happened: GitHub ended the preview on April 3, 2024 and said it would not become a product. The team’s work was folded into Microsoft’s VS Code Speech extension.
Use instead: VS Code Speech if you live in Copilot Chat, or a system-wide dictation tool if you want the same thing in every app and terminal.
Mozilla DeepSpeech — dead
What it was: The default open-source speech-to-text engine of the late 2010s. If a resource list from 2020 has an “open source STT” section, DeepSpeech is in it.
What happened: Mozilla stopped active development in 2021 and the repo was later archived. The team spun out to Coqui, which is the next grave down.
Use instead: The Whisper family. whisper.cpp for C/C++ embedding, faster-whisper for Python and GPU speed, Vosk if you need something tiny that runs on a Raspberry Pi. Here’s how they compare and how to run Whisper locally on Windows.
Coqui STT — dead
What it was: The continuation of DeepSpeech by the original Mozilla team, plus a well-regarded TTS line.
What happened: Coqui announced its shutdown in late 2023 and went offline in January 2024. The STT repo was archived.
Use instead: Same answer as DeepSpeech: Whisper won this category outright.
Dragon Professional for Mac — dead
What it was: Nuance’s flagship dictation suite on macOS.
What happened: Nuance discontinued the Mac product line in 2018 and ended support in 2021. Dragon is Windows-only now, and every “best dictation software for Mac” article that predates this still links to a product you cannot buy.
Use instead: On a Mac, SuperWhisper or MacWhisper, both local Whisper apps that outperform 2018 Dragon anyway.
Windows Speech Recognition — deprecated
What it was: The built-in Windows dictation and control layer that every “dictate on Windows for free” article recommended for fifteen years.
What happened: Microsoft deprecated WSR in December 2023. Voice Access replaces it on Windows 11.
Use instead: Voice Access for hands-free OS control. For dictating prose and prompts, anything Whisper-based is dramatically more accurate; the Windows options are here.
Serenade — stalled
What it was: The most polished attempt at natural-language voice coding, with editor plugins and a real parser. Open-sourced in 2022.
What happened: Not officially dead, and that’s the trap. The site is up and the repos exist, but active development effectively ceased around 2022 and nobody is answering the issues. Don’t build a workflow on it in 2026.
Use instead: Depends which half of Serenade you wanted. For structured editing by voice: Talon plus Cursorless. For “say what you want and code appears”: that’s what AI agents do now, fed by any good dictation tool. My take on that workflow is in voice-to-text for Claude Code.
Const-me WhisperDesktop — stalled
What it was: A fast Windows GUI for Whisper with Direct3D GPU acceleration, briefly the easiest way to run Whisper on a PC.
What happened: Development stalled; no meaningful updates since 2023, issues pile up unanswered. It still runs, but it’s frozen with the original Whisper models and their known repetition bugs.
Use instead: A current Whisper desktop app, or faster-whisper directly if you want the engine without a GUI.
Invoke is $49 once. Free 7-day trial, all features, no credit card.
Smaller graves
| Tool | What it was | Status | Use instead |
|---|---|---|---|
| Silvius | Kaldi-based voice coding research demo (2016) | Dead | Talon |
| Aenea | Dragonfly-over-network for Linux voice coding | Dormant | Talon (native Linux support) |
| Caster | Command grammar layer on Dragonfly | Largely dormant, check the repo before depending on it | Talon |
| Simon | KDE’s open-source speech control project | Dead, unmaintained since ~2016 | Voice Access / Talon |
| CMU Sphinx / PocketSphinx | The pre-deep-learning open STT standard | Frozen, its own docs point newcomers elsewhere | Vosk or Whisper |
| KnowBrainer forums (old URLs) | The Dragon community’s hub | Moved, old knowbrainer.com/forums links are dead |
forums.knowbrainer.com |
| “Hey GitHub” waitlist pages | Copilot Voice signup links | Dead | See the Copilot Voice entry above |
What to actually use, by situation
Your hands are the problem (RSI, injury, disability). Talon plus Cursorless, full stop. Budget real learning time, lean on the community Slack, and read handsfreecoding.org. This is the one case where the grammar approach is still the right answer, because you need to control everything, not just enter text.
You type fine but talk faster, and you drive AI coding tools. You want push-to-talk dictation with an AI reformatter, not grammars. This is what I built Invoke for, hold a key, describe the feature, release, and a clean prompt lands in Claude Code or Cursor. Here’s what a day of that looks like. On Mac, SuperWhisper is the strong local option.
You’re building software that needs speech-to-text. faster-whisper on a GPU for accuracy and speed, whisper.cpp for portability, Vosk for tiny devices. Skip everything with a grave above, no matter how many tutorials link to it.
Corrections
I check every status on this page against the source repos and vendor sites. Last verified: August 5, 2026. If something here has died, revived, or shipped since, email zach@scrn.co and I’ll fix it, this page is meant to be the one that stays accurate.
Disclosure: Invoke is my product. Everything else on this page is judged on its merits, including the competitors I link to, because a resource page that hides the landscape isn’t a resource page. Invoke is $49 once, runs Whisper locally on your GPU, and has a free 7-day trial, no credit card.