← Portfolio

Voice Transcript Agent · AI & Workflow

Local audio transcription desktop app

Capture audio, transcribe it locally and turn the recording into a usable record.

Explore the product
The product interface view

At a glance

What we built

A desktop workflow for recording audio, processing it locally and producing timestamped transcripts and structured output files, with optional language-conversion steps.

Project overview
ProductVoice Transcript Agent
Interfaces shownDesktop / web
TechnologyElectron · React · Swift · Python · Whisper
Core workMicrophone, system and mixed audio paths · Local Whisper transcription pipeline · TXT, SRT and JSON output workflow

Product flow

How it works

01

Microphone, system and mixed audio paths

02

Local Whisper transcription pipeline

03

TXT, SRT and JSON output workflow

The interface

Desktop interface · 01

Microphone, system and mixed audio paths

Design notes

Decisions behind the experience

The recording task

One desktop place for the audio and the words

  • Microphone and system-audio capture
  • Recording folders with session metadata
Read the context

A meeting, interview or spoken note becomes useful when its recording and transcript can be found together. The desktop app starts with a choice of microphone, system audio or both, then keeps the recording session visible through capture and transcription.

The mixed mode matters for calls: the microphone and computer output are recorded as separate inputs before they are combined. A single recording folder holds the original audio, metadata and every resulting text format.

From sound to transcript

Process speech locally and preserve timing

  • Local Whisper processing
  • Timestamped transcript segments
Read the context

The application normalises the audio and passes it to a local Whisper worker. Each transcript segment retains timestamps, so a reader can move between the written words and the moment they were spoken.

Language detection is part of the transcription pass. The base workflow produces the original transcript without requiring a remote text service; optional English and Romanized variants are separate conversion steps.

Working with the result

Make the output useful beyond the app

  • TXT, SRT and JSON exports
  • Recent-session access
Read the context

A recording can leave the desktop as plain text, subtitles or structured JSON. These are different uses of the same session: text for reading and editing, SRT for timed playback, and JSON for software that needs segment data.

Recent recordings and their files remain accessible from the app. That keeps correction and reuse close to the original instead of scattering exports across unrelated folders.

Engineering

Join native capture to a familiar interface

  • Electron and React
  • Swift, FFmpeg, Python and Whisper
Read the context

Electron and React provide the desktop workspace. A Swift ScreenCaptureKit helper handles macOS system audio; FFmpeg prepares the audio; Python runs the local transcription worker. The application coordinates those stages and records errors in the session metadata.

This separation makes the interface simple without hiding the work. Capture, processing and export each have a distinct state, while the files remain available for a later review.

Design and development

What powers it

Platform and technology

Electron · React · Swift · Python · Whisper

Core capabilities

  • Microphone, system and mixed audio paths
  • Local Whisper transcription pipeline
  • TXT, SRT and JSON output workflow

Questions about the product

Questions about this product

What does this product help people do?

Capture audio, transcribe it locally and turn the recording into a usable record.

Which screens can I explore?

Microphone, system and mixed audio paths.

What capabilities are covered?

Microphone, system and mixed audio paths, Local Whisper transcription pipeline, TXT, SRT and JSON output workflow.

AI & Workflow projects

More work to explore

View all projects ↗