Annotate AudioDownload

On-device sound classifier

Teach your
iPhone
any sound. Annotate Audio: train a machine-learning sound classifier on your iPhone.

Record examples, train a classifier on your iPhone, let it listen.

  • Machine learning
  • Create ML
  • Core ML
  • On device
Download on the App StoreiPhone · iOS 18 or later
Annotate Audio auto-recording a bark it recognised with 97 % confidence
Listening

A live meter playing real recordings of a bark, a snore, a cough, a horn and a sneeze, tagging each sound as it is recognised.

How it works

Label. Record. Train. Detect.

Machine learning for sound, without code or cloud.

  1. The recordings list: one folder per sound, each with its level

    01

    Label

    Each sound is a class with its own folder. Add a “Not …” class so it learns the difference.

  2. Recording a sound on the record screen with a live waveform

    02

    Record

    Every recording is a training example. Five per class is enough to start.

  3. Training a sound classifier on the iPhone, with progress

    03

    Train

    Tap Train. Create ML builds your sound classifier on the iPhone in a few minutes.

  4. The trained model recognising a bark and saving the clip

    04

    Detect

    Turn Auto on. The model listens live and saves a clip when it recognises a class.

Training data

From Puppy to Top dog.

  1. Puppy

    0

  2. Learning

    5

  3. Good dog

    50

  4. Top dog

    100

Hi! I learn sounds.

Reps

000

Puppy

Pre-buffer

Never miss the start.

Every auto clip starts 2 seconds before the model recognised the sound.

2 s before3 s after
Listening

Real clips from the app. The first barks came before the model was sure, and they are in the recording.

Recording by hand? Turn on the pre-buffer and your clip starts 2 seconds before you tapped.

The Barking folder: every clip with its waveform, ready to play

Made for real life

Only the sound you asked for.

Train a class for any sound, leave it listening, and come back to just those clips.

  • BarkingBarking or whining while you're away.
  • SnoringSnoring and sleep talking at night.
  • CoughingA cough in the night, clip by clip.
  • HonkingHorns, alarms and doorbells.
  • Running waterAppliances and machines.
  • TypingAny sound you want to catch, count or collect.

Details

A real ML tool, in your pocket.

Multi-class models

One model knows several sounds. Each clip lands in the folder of its class.

Confidence threshold

Set how sure the model must be before it records: 85, 90 or 95 %.

Model report

After training you see its accuracy, and clips it disagrees with are flagged for review.

Clean training data

Move a mislabelled clip to the right class, then train again.

Dataset export

ZIP any folder, with a training layout for Create ML on Mac.

Listens for hours

Keeps listening in the background and dims the screen when the phone lies face down.

Private by design

Your sounds stay yours.

Your data stays on your devices and in your own iCloud.

  • Trained on your iPhone

    Create ML trains the model and Core ML runs it, both on your device. No cloud, no server.

  • Your own iCloud

    Recordings and models sync through your iCloud Drive, not a server of ours.

  • No account

    No sign-up, no login, no profile.

Read the privacy policy →

FAQ

Good questions.

Something else? Get in touch

Your own sound classifier

Teach it your first sound.

Five examples of two sounds is all it takes to train.

Download on the App Store