Patterns we can look for
Barks, yips, howls, growls, whimpers, meows, purrs, hisses and caterwauls. Quiet, mixed or unfamiliar sounds may be missed.
LISTEN. NOTICE. CONNECT.
Listen closer.
Connect better.
Explore your dog or cat’s sounds.
Try your words in a real animal voice.
Audio and typed requests stay on your deviceSound clues, with honest limits
01
Let’s hear your pet
ExperimentalEvery sound starts a story.
Record 5–10 seconds in a quiet space.
02Add what you can seeOptional, but helpful
Context helps explain possibilities. It is not measured by the audio model.
Your recording stays on your device.
First analysis downloads a large model, about 90 MB plus runtime files. Wi-Fi recommended. No account or payment needed.
YOUR WORDS → AN ANIMAL VOICE
Start with play.
Hear a real dog.
Write a short invitation to play. A multilingual language model looks for the intention, then selects a real dog vocalisation recorded during play.
“มาเล่นกันเถอะ” Play Natural dog sound
This is our first text-to-animal-sound experiment. The recording’s play context is documented by its creator. Whether another dog hears it as an invitation is still untested.
Try “Let’s play” or “Lass uns spielen.” Only dog play is available. Other meanings and animals need their own evidence.
Your text stays on this device. First use downloads about 118 MB plus runtime files.
Already have a cue your pet knows? Use your own recording
YOUR WORDS → THEIR LEARNED SOUND
Your voice.
A real connection.
Try a practical communication experiment. Record a word, whistle or other cue your dog or cat already knows. A multilingual language model can match your request to that real recording.
The model interprets human words. The connection to your pet comes from their learning. We cannot turn an arbitrary sentence into meaningful barks or meows.
“มานี่” Come Your recorded recall cue
No generated animal sounds. No voice synthesis. Use your own recordings, without provoking an animal. Keep volume low and stop if your pet seems uncomfortable.
1. Add a real, familiar cue
Use recordings for one pet at a time. They stay in this tab and disappear on reload.
2. Ask in your language
One intention at a time: come, sit, wait, play, food, outside. Always check the model’s suggestion.
Runs on your device. First use downloads about 120 MB plus runtime files. Nothing plays automatically.
WHAT WE CAN KNOW
A sound recogniser, with an honest boundary.
Pet Listen uses an AudioSet model to look for dog and cat vocalisations, then offers cautious context-based possibilities. A separate multilingual text model matches simple human requests to intentions. The animal-voice experiment retrieves a real dog recording from a reported play context. It does not yet establish that dogs understand the intended message. Familiar learned cues remain available separately. Neither model translates your pet’s thoughts, confirms an emotion, or diagnoses illness.
Context matters
The same call can happen in different situations. Food nearby does not prove hunger; purring does not prove a cat is happy or well.
Still to be established
This product has not been validated on a representative set of pets. Cautious screening rules reduce guessing, but do not guarantee correctness.
Research and model sources
- Multilingual MiniLM sentence model: matches human-language requests to a small cue vocabulary. This is semantic similarity, not a dog or cat language model. Every suggestion needs your confirmation.
- MIT Audio Spectrogram Transformer, with the Transformers.js ONNX conversion. A sound event model, not an emotion model.
- Google AudioSet: human-labelled sound events in short clips.
- What’s in a Meow? (2020): human listeners had limited ability to identify the recorded context of cat meows.
- Towards Dog Bark Decoding (2024): research into bark classification, not a universal dog language.
Privacy and recording help
Audio is processed in your browser and is not uploaded by Pet Listen. It is held in this tab until removed or closed. The model and runtime download from Hugging Face and its delivery services, which receive ordinary network metadata such as your IP address. The model may be cached by your browser. Clearing this site’s storage and cache removes cached files.
Microphone access starts only when you press record and stops after 30 seconds, when you stop, or when this tab is hidden. Select a short WAV, MP3 or M4A if live recording is unavailable. Formats depend on your browser. A quiet recording may need to be made closer to your pet.
Pet Listen does not collect an account, payment details or an audio library. MagicAds displays sponsored placements and receives ordinary request metadata, the page address and public page information for ad delivery and impression/click measurement. Your recording, filename, typed request, familiar cue, response notes, selected context and sound result are not sent to MagicAds. Hosting providers may retain standard access logs. A saved sound observation downloads to your device and contains your chosen context and model results, but no audio. Animal-voice requests, familiar-cue requests and response notes remain in this tab; they are not saved by Pet Listen or used to train a model. Familiar cue recordings stay in this tab and disappear on reload. They are played as recorded, with no synthetic bark or voice generation. Cue microphone recording stops after 10 seconds, on cancellation, or when the tab is hidden. Generated animal artwork is illustrative.
Use with care
This experimental tool is for observation and education. Never provoke an animal to test it or rely on it to decide whether an animal is safe to approach. If a pet has trouble breathing, collapses, appears in severe pain or seems acutely unwell, seek urgent veterinary care regardless of any result.
Errors and missed sounds are possible. Pet Listen provides no veterinary diagnosis or guarantee of accuracy. Use the “That sound seems wrong” control to flag a result in your observation; it does not send a message or improve the model automatically.