“Nobody talks like that!” – Experiences with AI in text, images and speech
Hearonymus has been exploring the possibilities of artificial intelligence right from the start. We are constantly building up our expertise, particularly in areas directly related to the production of audio guides. The key areas in which we regularly assess quality are: text creation (editing), audio production from text, and the creation of visual material. Our assessment in brief:
Text creation: We asked ChatGPT to create a “walking tour of the ten most important sights in a specific small town”. The result was interesting. At least the number of suggested sights was correct – ten.
The errors then surfaced in the details: An inhabited castle was turned into a ruin by the AI. A charnel house that doesn’t exist was recommended as a tourist attraction. A nature park 20 kilometers from the small town was also suggested. Likewise, a castle located 10 kilometers from the town, in the opposite direction to the nature park, was also suggested. This imaginary tour is not realistically manageable on foot.
Whilst the texts produced are quite well written, some of the content is simply incorrect. Hearonymus therefore continues to rely on human editors with a solid educational background, who specialise in specific topics, carry out thorough research and have an innate understanding of the difference between a text intended for reading and one intended for listening.
Audio: AI’s ability to generate audio from written text is gradually improving. Italian and English already sound reasonably natural, but there is still a great deal of work ahead for AI developers when it comes to German. In any case, the intonation of professional voice-over artists is different from that of machine-generated speech. We will continue to monitor developments and keep you up to date.
Image: The advances in the field of ‘image generation’ are simply incredible. Over the past few months, AI has improved to such an extent that it is now possible to have almost any image created digitally. For all businesses that use images in their day-to-day work, this is wonderful news and a real relief. For businesses that make a living from producing and selling images, however, this news is not welcome. It is not only the business models of image databases and (advertising) photographers that will feel the full brunt of this in the near future.
Overall, when it comes to writing texts and producing audio, we continue to work with our editors, professional voice-over artists and our recording studios to ensure the highest quality. You can really hear the difference. When it comes to creating images, we are happy to make use of AI. All the images in this newsletter – except those in the article ‘New Guides’ – were created using AI.
Picture (c) Created with artificial intelligence (AI)