Live Farsi Audio Transcription Tool
A web app that transcribes Farsi (Persian) audio files to text — deployed live and handling real-world file sizes through intelligent chunked processing.
Farsi Transcriber was built to solve a practical problem: accurately converting spoken Farsi audio into clean, readable text. The app leverages OpenAI's gpt-4o-transcribe model — one of the most capable multilingual speech recognition models available — and wraps it in a clean, accessible Streamlit interface that anyone can use without technical knowledge.
A core challenge with audio transcription is handling large files reliably. Rather than attempting to upload multi-hundred-megabyte files directly to the API, the app uses pydub to split audio into 200-second chunks and transcribes them sequentially, showing live progress as each segment completes. This makes the tool robust even for long recordings like lectures or interviews.
Technical Implementation
Architecture
- UI Layer: Streamlit app managing multi-step session state (API setup → upload → transcription → results)
- Core Logic: Dedicated
api/transcriber.pymodule handling chunking and OpenAI calls - Audio Processing: pydub + FFmpeg for format conversion and segmentation
- Deployment: Containerized via Docker, live at farsi-transcriber.jonahsaidian.com
Processing Pipeline
- User uploads audio file via drag-and-drop
- File loaded in a background thread with progress estimation
chunk_audio()splits into 200-second WAV segments- Each segment sent to OpenAI with
language="fa" - Transcribed text accumulates live in the UI
- Temp files cleaned up after each chunk
Features
-
File Upload
Drag-and-drop support for MP3, MP4, MPEG, MPGA, M4A, WAV, and WEBM formats.
-
Real-Time Progress
Live progress bar with time estimates and streaming transcription display as each chunk completes.
-
Text Export
Download the full transcription as a
.txtfile with one click.
-
Chunked Processing
Handles large audio files reliably by processing in 200-second segments — no file size bottlenecks.
-
Auto API Key
Reads
OPENAI_API_KEYfrom environment or.env, with a fallback UI input field. -
Clean Reset
One-click restart clears all state and the file uploader, ready for a new transcription.
Live Deployment
The app is publicly accessible at farsi-transcriber.jonahsaidian.com — built with OpenAI gpt-4o-transcribe for state-of-the-art Farsi speech recognition.