Home → Blog → How to turn voice notes into text automatically

How to turn voice notes into text automatically

Published 2026-08-20 · 5 min read

Ideas arrive on a walk, and talking is faster than typing while moving. The problem shows up later: a folder with a hundred recordings and no way to find anything in them. Transcription turns that pile into ordinary text you can search and link.

A pipeline with no manual steps

The working setup has three parts:

The recorder app on your phone writes into a folder that syncs to your computer.
A script on the computer notices new files and sends them for transcription.
The text lands as a note next to the original audio.

Nothing to click: you speak, and a minute later the note is in your knowledge base.

The watcher script

The simplest version needs no dependencies — walk the folder and process whatever has no text yet.

import pathlib, datetime
from openai import OpenAI

client = OpenAI(base_url="https://voicesscribe.com/v1", api_key="your key")
vault = pathlib.Path.home() / "Vault" / "voice"

for audio in vault.glob("*.m4a"):
    note = audio.with_suffix(".md")
    if note.exists():
        continue
    with audio.open("rb") as f:
        text = client.audio.transcriptions.create(model="whisper-1", file=f).text
    stamp = datetime.datetime.fromtimestamp(audio.stat().st_mtime)
    note.write_text(
        f"---\ndate: {stamp:%Y-%m-%d %H:%M}\nsource: {audio.name}\n---\n\n{text}\n",
        encoding="utf-8")
    print("done:", note.name)

Schedule it every five minutes and notes appear practically as you record them.

Give the note a title and shape

A raw transcript is one long stream of speech. To be useful a note needs a heading and at least rough structure, and that is where a language model helps: send the same text for a summary and put the result at the top as an overview plus a task list.

Keep the full transcript too. A summary drops details, and you will search your knowledge base using the exact phrases you said out loud.

Accuracy on field recordings

Notes on the move are hard material: wind, footsteps, swallowed endings. Three habits that visibly improve the result:

Hold the phone closer to your mouth — it matters more than any setting.
Pass the language code if you always dictate in the same one.
Keep a standing prompt with project names, colleagues and domain terms.

Try it on your own recordings. Sign-up takes a minute, and the free minutes are enough to judge the quality.

Get a free API key

Frequently asked questions

What format do phone recorders produce?

Usually m4a or ogg, both accepted as they are. No conversion step is required.

What about notes in two languages?

The language is detected from the audio, so a mixed archive needs no configuration. Set the language explicitly only if the stream is reliably monolingual.

Can this run without a computer?

Yes, if your automation platform can watch a cloud folder — then the whole flow lives in the cloud.

What happens to the audio afterwards?

Recordings and transcripts are kept no longer than 24 hours and then deleted automatically.