Google researchers, demonstrating a new AI-based translation model, have shown how people may soon be able to use AI to seamlessly speak to other people in different languages using their own voice. — Photo: Google/dpa
BERLIN: AI is ushering in a new generation of translation apps, and Google's latest translation project is trying to break down language barriers with a new AI model that does the talking for you.
Google's Audio-Palm AI model for translation, unveiled in late June, can recognise, process and generate both text and speech, but the most impressive feature is that it can do it with the respective user's own voice.
The scientists have demonstrated the results of their work in a with a dialogue between several people speaking different languages, with their voices all seamlessly translated into spoken English.
Google Translate has fallen behind rival translation tools like DeepL in terms of quality, but the developers say the new model "significantly outperforms existing systems for speech translation."
The user also doesn't need to spend much time training the AI translator to speak like them, and the developers say only a "short spoken prompt" is needed.
Audio-Palm is the combination of the AI language model Palm with the AI audio generator Audio-LM. Palm is also used in Google's chatbot Bard.
While the concept lends itself to visions of people at parties or workplaces being able to effortlessly speak various different languages to each other, users would presumably still need to finish speaking before the translation could begin. There's also no indication yet as to if and when Audio-Palm will be used in Google Translate. – dpa
