AI Tooling

TinyTalk

Right-click transcription with speaker detection. Local Whisper — no API key, no cloud, no console window.

Select any audio or video file, right-click, transcribe. faster-whisper runs entirely offline, labels multiple speakers automatically, and streams timestamped output live. Saves .txt next to source.

Download v1.3.1 (.exe)GitHub →

Windows 10/11 · No admin rights required · ~150MB model download on first use

How It Works

TinyTalk registers a context menu entry for all supported audio and video formats. When triggered, it launches a headless faster-whisper process that streams transcript segments to a lightweight output window as they arrive — no waiting for the whole file to finish.

On first run it downloads the base Whisper model (~150MB) automatically. Subsequent model updates happen in the background while the current model stays active.

Supported Formats

.mp3.wav.m4a.flac.ogg.aac.mp4.mkv.mov.webm.avi.wma.opus.mp2.mts

Model Sizes

tinyFastest, lowest accuracy

base ✦Default — good balance of speed and accuracy

smallBetter accuracy, slightly slower

mediumHigh accuracy, requires more RAM

large-v3Best accuracy, slowest

✦ Default model — auto-downloaded on first use

Features

Speaker Detection

Transcript automatically labels SPEAKER 1 / SPEAKER 2 etc. when multiple voices are detected. No config, no API keys — runs entirely on-device.

Per-Sentence Timestamps

Every clause gets its own line with a [M:SS] prefix in both the live log and the saved .txt. Easy to scan, easy to scrub to.

Fully Offline

No API key, no internet connection, no cloud processing. All transcription runs locally via faster-whisper. Your audio never leaves your machine.

Right-Click Anywhere

Select any audio or video file in Windows Explorer, right-click, and hit Transcribe. Works on single files and batches.

ETA During Transcription

Processing speed is measured in real time and used to estimate remaining time. No more staring at a progress bar with no idea when it'll finish.

Live Streaming Transcript

Output streams segment by segment as Whisper processes — you see the transcript building in real time rather than waiting for the whole file.

Auto-Downloads Model

First run pulls the ~150MB base model automatically. Background model updates download silently while the current one stays active — no restarts.

No Admin Rights

All registry entries written to HKCU. Installs without an elevation prompt. Context menu registered for 15 audio and video formats.

Install

One-click (recommended)

Download TinyTalkSetup.exe from Releases and run it. Installs to %LOCALAPPDATA%\TinyTalk\ and registers all context menu entries — no admin prompt.

Manual install

powershell -ExecutionPolicy Bypass -File install.ps1

Uninstall

powershell -ExecutionPolicy Bypass -File "%LOCALAPPDATA%\TinyTalk\uninstall.ps1"

Stack

Python 3.8+faster-whisperOpenAI WhisperPowerShellHKCU RegistryIExpress Installer