Local hold-to-talk dictation for secure on-device transcription
Dictate into any focused application by holding a hotkey: press, speak, release and the text is typed where the cursor sits. Mohsen Seyedkazemi Ardebili built YazSes as a background daemon utilities tool that runs locally on desktop machines. It uses a local faster-whisper engine, supports file transcription and meeting capture, and offers regex-based voice commands. The app fits privacy-conscious professionals and developers working in air-gapped or sensitive environments who need offline speech-to-text.
What the tool does and how it injects text
YazSes captures audio on demand and injects transcription into the active window. The app implements a hold-to-talk hotkey for live dictation and a hands-free meeting mode for continuous capture. It also accepts audio files for batch transcription into TXT, MD, SRT, VTT, or JSON. The design targets situations where you need text typed directly into editors, terminals, or browsers without switching workflows.
How it affects system resources during transcription
Transcription runs locally on the CPU and has measurable memory needs. The app uses faster-whisper with int8 quantization so it does not require a GPU, but the documented minimum is 4GB RAM and 8GB is recommended for smoother operation. The processing model executes on-device, which keeps latency low for real-time dictation on standard consumer CPUs while using CPU cycles during active transcription.
Safety, privacy, and operational guarantees
All processing stays on the machine and the app collects no telemetry. YazSes operates without streaming audio to external services, and it requires no accounts or API keys. For sensitive environments this means no audio or text data leaves the host. Optional advanced features may rely on locally installed models, so administrators must provide those artifacts rather than sending data offsite.
Who can run it and how much technical setup it needs
The tool expects some comfort with package managers and local configuration. Windows installation is available through the Windows Package Manager command, and macOS or Linux installs use Homebrew, Snap, or pipx. Voice macros use a regex-based grammar, which offers power for developers but requires learning the syntax. The app targets users who can handle modest command-line steps and local model management.
Practical local dictation for privacy-focused users, with modest setup trade-offs
YazSes is a focused option for professionals who require offline speech-to-text and direct text injection into applications; its main trade-off is a need for modest RAM and occasional command-line setup to install or swap local models. For anyone operating in air-gapped or sensitive environments, the app provides predictable, on-device transcription without relying on third-party services.
