Focus a field
Click where the dictated text should go. Blabbermouth remembers the Windows app and control.
✓ target: editable_control
Windows 10/11 · x64 · v1.9.2
Speak anywhere.
Type everywhere.
Private, low-latency speech-to-text for Windows. Focus a field, hold a shortcut, speak, release, and Blabbermouth delivers the finished text without sending your audio to a cloud transcription service.
No account. Optional local models. Designed for editable Windows fields.
Ctrl+Shiftsilent modifier-only
activation
activation
PS C:\Users\you> git status
On branch main
nothing to commit, working tree clean
PS C:\Users\you>
PS C:\Users\you>
LISTENING · EMAIL CAPTUREDRelease Ctrl + Shift to transcribe
TOSam Carter
SUBJECTRelease notes
The Windows build is ready for review. I’ve attached the final notes.
Local processing
Works across Windows apps
Choose your model
No account required
01 / HOT PATH
One shortcut.
Zero context switches.
Blabbermouth captures the field you were using, keeps recording state obvious, then returns the completed result when you finish.
Hold and speak
Use hold-to-talk, toggle mode, or a silent modifier-only shortcut. The topmost overlay confirms recording.
Release. Continue.
Your selected local model transcribes, applies enabled text rules, and delivers the finished text.
CtrlShift↑
02 / DATA PATH
Local is the architecture.
Normal dictation stays on your PC. No account, no cloud transcription, and no microphone-audio history.
Microphone capture
Audio is captured for the current recording. The last completed WAV remains memory-only for recovery unless you explicitly save it.
Your chosen model
Whisper or Parakeet processes speech on this Windows PC. Download model files only after installation.
Targeted delivery
Finished text is sent back to the captured editable field using Unicode keystrokes or a native edit-control path.
03 / TOOLKIT
Dictation tuned like a tool.
Build a workflow around your vocabulary, your applications, and the way you write.
Use the shortcut that stays out of your way.
Choose Ctrl, Alt, Shift and Win combinations, with an optional primary key. Modifier-only combinations are silent in terminals; keyed shortcuts suppress their printable key while active.
HOLD-TO-TALKTOGGLE MODESILENCE COMPLETIONESC TO CANCEL
Change tone by app.
Set formal, casual, or verbatim output profiles by destination process.
Teach Whisper your terms.
Condition Whisper with product names, proper nouns, and specialist vocabulary.
Correct recurring misses.
Map heard phrases to fixed corrections. Keep every rule visible and editable.
Speak reusable snippets.
Expand exact triggers with dates, times, new lines, and tabs inside sentences.
Punctuation on command.
Optionally convert spoken punctuation, remove isolated filler words, and append a trailing space.
04 / MODEL CATALOGUE
Pick the engine.
See the weight.
The installer stays lean. Choose and download a local model after installation, with storage needs shown before selection.
| Model | Download | Language scope | Use |
|---|---|---|---|
| Whisper Small | ~466 MiB | Multilingual | Lighter Whisper option for everyday local dictation |
| Whisper Medium | ~1.5 GiB | Multilingual | Larger Whisper option for users who favour accuracy |
| Whisper Large v3 | ~2.9 GiB | Multilingual | Largest included Whisper catalogue choice |
| Parakeet TDT 0.6B v3 INT8 | ~465 MiB | 25 European languages | Fast local alternative powered by sherpa-onnx |
Whisper supports optional translation into English and NVIDIA CUDA where supported, with CPU fallback. Parakeet currently runs on CPU and does not provide translation. Compatible ggml Whisper models can also be imported from disk.
05 / AFTER CAPTURE
History for text.
A queue for media.
Recover useful words without silently building an audio archive.
Search, favourite, correct, retype.
Optionally retain the 50 most recent text transcripts locally. Export filtered history as Markdown or JSON, and turn an unambiguous one-word correction into a visible replacement rule.
Microphone audio is not added to history.
batch_queue / 03 files
interview-01.m4aDONE · SRT
meeting-notes.mp3DONE · TXT
demo-capture.mp4PROCESSING
WAV · MP3 · M4A · AAC · WMA · MP4 · MOV
Optional .transcript.txt, SRT, and WebVTT exports.
06 / OPERATING BOUNDARIES
Exact claims.
No hand-waving.
Compatibility depends on the destination control and Windows integrity level. Here is the honest edge of the current product.
01
Windows 10 and Windows 11, x64 desktop only.
02
Text arrives after recording finishes; live word-by-word typing is not shipped.
03
Dictating into an elevated target requires Blabbermouth to run elevated too.
04
Designed for editable Windows fields. Compatibility is not guaranteed in every application or control.
05
A speech model must be downloaded before offline recognition can work.
READY WHEN YOUR FIELD IS
Keep your hands on the work.
Install Blabbermouth, choose a local model, then use one shortcut anywhere you write on Windows.
[OFFICIAL DOWNLOAD URL REQUIRED] · v1.9.2 · Windows 10/11 x64