About Services Tools Blog Contact us

Audio File Transcription

Turns a spoken audio file into text with Whisper, and can work through product voice notes that have no transcript yet.

  • Speech Recognition
  • Footwear retail
  • Scheduled worker
  • Productised
  • Hybrid
  • 2025

The problem

Typing up a recording by hand delays captions, notes and product copy. Spoken fit advice cannot be searched or reused in a description until it is text.

What it does

A command takes an audio file, a place to save and a model size, and writes plain text without speaker labels or timestamps. A scheduled run uses a larger model, picks product recordings with an empty transcript, fetches each file and saves the English text on that record. If the database connection fails, the transcript stays empty so the next run can retry it.

Outcomes

  • Each audio file gets a transcript beside it without a paid speech service.
  • Product voice notes become searchable text on the record.
  • A failed connection leaves the recording on the waiting list instead of losing it.
  • The model size can be a fast draft or a more careful pass.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Transcribe a file

Audio
Click to load the sample voice note

MP3, WAV or M4A. The stored file is plain text with no speaker labels. The player follows segment times from this pass.

Model and run

A short note finishes in a few seconds. Useful for a first pass.

Default for a single file. English only, plain text out.

Load the sample, then transcribe
Recent files
FileTranscriptModelStatus
court-note-navy.m4a A soft leather court with a low block heel. Balanced written
boot-note-black.wav Elastic sides, a stacked heel, and a black finish. Fast draft written

Product voice notes

Empty transcripts
0
Already stored
0
Queue model
Large
English
This pass
One recording is still blank
Recordings
IdProductTranscriptStatus
vn-1044 Harbour court, navy A soft leather court with a low block heel. stored
vn-1048 Mill ankle boot, black Elastic sides, a stacked heel, and a black finish. stored
vn-1051 Stone trainer, wide (empty) waiting

Failed connection

Connection refused The recording stays on the row. The transcript is left empty so the next run can retry.
Last error
Down
Still empty
0
Recording kept
Yes
Attempt
Ready to replay the failed pass
vn-1051 · Stone trainer, wide
AudioTranscriptNext step
stone-trainer-wide.m4a (empty) Retry

How it works

  1. Accept a local audio file and a model size, or pick product records with an empty transcript.
  2. Fetch hosted recordings when there is no local copy.
  3. Run Whisper in English.
  4. Write a plain-text file, or save the text on the recording's record.
  5. Leave a failed record empty so the next run tries again.

Pipeline

flowchart LR
  Audio[Audio file] --> Model[Whisper]
  Queue[Empty transcripts] --> Model
  Model --> Text[Plain transcript]
  Text --> File[Text file]
  Text --> Row[Product row]

Sample output

Wide-fit trainer note in a British voice, with the local transcript.