NoTrack Coder 2.9.41 — talk to your agent, and have it read back, all on your PC
NoTrack Coder 2.9.41 lets you talk to your agent and listen to it talk back — without a single byte leaving your PC. Two new things: you can dictate a message instead of typing it, and you can have any message read aloud. Both the listening and the speaking happen locally on your own machine, in keeping with the whole point of NoTrack Coder: your work stays with you.
Speak instead of type
There's now a microphone icon in the message box, right next to Send. Click it and start talking; click it again when you're done. The recognized text drops straight into the message box at your cursor — you read it over, edit if you like, and send it yourself. Nothing is sent on your behalf.
Speak in any language — it's detected automatically, so there's nothing to set. It's handy when your hands are busy, when a long thought is faster said than typed, or when dictating simply fits how you think.
Hear any message read back
Every message — both yours and the agent's — now has a speaker icon in the small icon bar beneath it. Click it to listen, click again to stop. Code blocks, links, tables and emoji are skipped, so you hear the actual prose and not a stream of punctuation. Reading begins as soon as the first sentence is ready, so even a long reply starts playing right away instead of making you wait.
Your voice and your text stay on the machine
This is the part that matters most. The speech recognition runs locally, on your own computer — the audio from your microphone is never uploaded anywhere. The reading-aloud is generated locally too — the text of your messages is never sent off the machine to be turned into speech. No cloud speech service, no account hand-off, nothing phoning home. Dictation is handled on-device by Whisper, and the spoken voice by Piper — both open, both running entirely on your PC.
Downloaded only when you ask, removable anytime
Nothing for voice is installed until you first use it. The first time you click the microphone, the app checks your PC and asks for your consent before downloading the recognition model from Hugging Face — the larger, more accurate model (about 1.2 GB) on computers with an NVIDIA graphics card, or a lighter one (about 0.5 GB) otherwise. Progress shows right on the microphone icon. If a computer is too limited for it (under 4 GB of RAM, or not enough free disk), the app tells you plainly instead of failing quietly.
Read aloud works the same way: the first time you use it, the app asks to install a small speech engine (about 20 MB) plus a voice for the language of that message (about 60–75 MB). Each new language pulls its own voice the first time you read something in it, always after you agree.
Everything that gets installed is listed under Settings → Local Models → Voice input / Read aloud, where you can see exactly what's on disk and remove any of it whenever you want.
Available on Standard and Pro
Voice input and read aloud are part of the Standard and Pro plans, on Windows and Linux (x64). Free users see the icons too — clicking one explains that it's included with Standard and Pro.
Get 2.9.41
Download for Windows, or let your install auto-update. Windows 10 and 11. On Linux? The Linux build is right here.
Frequently asked questions
What's new in NoTrack Coder 2.9.41?
Two features, both fully on-device. Voice input: a microphone button in the message box lets you dictate instead of type — click it, speak, click again, and the recognized text appears at your cursor for you to review and send. Read aloud: a speaker button under every message plays it back, and stops when you click again. The language is detected automatically in both, and neither your audio nor your text ever leaves your computer.
Does my voice or my text leave my computer?
No. Speech recognition for voice input runs locally on your PC (by Whisper), so the microphone audio is never uploaded. Read aloud is generated locally too (by Piper), so the message text is never sent off the machine. There's no cloud speech service involved in either.
What gets downloaded, and how big is it?
Nothing until you first use each feature, and only after you consent. Voice input downloads a recognition model on first use — about 1.2 GB on a PC with an NVIDIA graphics card, or about 0.5 GB otherwise. Read aloud installs a small speech engine (about 20 MB) plus a voice for the message's language (about 60–75 MB), and each additional language adds its own voice the first time. Everything is listed and removable under Settings → Local Models.
Which plans and systems support voice input and read aloud?
They're part of the Standard and Pro plans, on Windows and Linux (x64). Free users can see the icons; clicking one explains that the feature is included with Standard and Pro.