Wav2vec2: Artificial Intelligence with Dog Understanding: Bid Farewell to Not Understanding Your Dog's Barking Behavior
They say a dog is man’s best friend, so it makes sense that we would want to understand their language. Many AI systems are working on this, but one created by the University of Michigan is special because Wav2vec2 is a generative AI that was created to transcribe human language, not dog language...
There are dozens of AI systems that transcribe audio to text, meaning they can create subtitles from any dialogue they hear. The most famous is OpenAI’s Whisper, but there are others, such as Wav2vec2.
This type of AI is trained on human conversations so they can understand any transformation and transcribe it into text. But what happens if we train it on dog barking?
- Artificial intelligence that understands dog language
Researchers from the University of Michigan and Mexico's National Institute of Astrophysics, Optics, and Electronics have published a study that attempts to determine whether artificial intelligence can understand dogs' emotions from their barks.
They used the Wav2Vec2 model to recognize human speech, which they trained using the barks and vocalizations of 72 dogs of different breeds.
Recording dogs is almost more expensive than training AI. It's not easy to record different types of barking from the same dog because you don't know when it will bark, and you have to wait for it to happen.
To collect the data needed to train the language model, the researchers elicited some reactions, for example, by repeatedly ringing a bell or talking to them affectionately, to record the sounds they made.
Wav2Vec2 was able to detect the correct meaning of a bark 70% of the time. It is particularly good at distinguishing between barks that sound very similar, but some are aggressive and others are warning barks.
Not only that. From the dog's bark, you can also guess its breed, age, and gender, with the same success rate.
It won’t be long before we can install an AI-powered dog bark translator app on our mobile phone. What’s interesting about this experiment is that it relied on AI that was created to understand human language, not dog language.