All projects
Artificial Intelligence

Voice Transcript App

A transcription tool that turns audio into a Word document

localhost
Voice Transcript App
Streamlit
Interface
.docx
Output
2024
Year

Overview

A meeting recording, a lecture, an interview note: the moment audio files pile up, the information inside them becomes practically inaccessible. Nobody replays a two-hour recording from the start to find a single sentence.

The application solves that by the shortest route. The user drags and drops an audio file, a speech recognition model produces the text, and the result downloads directly as a formatted Word document. The Word output is a deliberate choice: in most organisations, post-transcription editing still happens in Word.

The interface is a single screen. There is nothing beyond upload, processing status and a download button — because the user of this tool might be a secretary or a student, not an AI practitioner.

Highlights

  • Drag-and-drop audio upload
  • Text generation via automatic speech recognition
  • Formatted Word (.docx) output
  • A minimal flow completed on one screen

Objectives

  • Turn information locked in recordings into searchable text
  • Offer an interface that needs no technical knowledge
  • Deliver output in the format organisations actually use

Solution

  • A single-page application flow with Streamlit
  • Integrating a speech recognition model
  • Generating formatted documents with python-docx
  • Chunked processing for long files

Approach

  1. 01Simplifying the use case
  2. 02Model selection and accuracy trials
  3. 03Testing the output format with users

Outcomes

  • Manual transcription work disappeared entirely
  • Output arrives in a directly editable format
  • Non-technical users can benefit without assistance

Challenges

  • Processing long audio files within memory limits
  • The effect of background noise on accuracy
  • Punctuation and capitalisation in Turkish recordings

Selected Work

Similar projects

View all