Ensemble pour le DAB+ urges French government to strengthen public support
Radio has adopted, if not fully embraced, AI — particularly the generative version embodied by ChatGPT — for various functions, including collating advertising data and creating virtual, voice-cloned presenters. Although AI has been around since the 1950s, it is ramping up again with new variations appearing that can surpass what generative AI has made possible.
Among the “new” AI types pushing the technology further are multimodal AI, agentic AI and customized enterprise forms of generative AI models. Although open AI sources have been available for some time, these are also now central to the ongoing growth of AI usage. Among the leading examples are Meta Llama 2 and Mistral Mixtral, both of which, like ChatGPT, offer large language model (LLM) capability using machine learning to handle massive amounts of data for natural language processing (NLP).
Multimodal AI systems comprise three modules: one for input, one for output and a “fusion” module. The input unit consists of multiple unimodal neural networks that receive a range of data formats. The fusion module combines, aligns and processes this data, and the output component delivers the result.
This form of AI can also detect patterns and connections when analyzing various data types, making it more accurate, intuitive and effective with a more contextual and human approach. ChatGPT developer OpenAI launched a multimodal AI, GPT-4o (“o” for “omni”), in May, claiming it has more natural human-computer interactivity for audio, video, text and images.
Proactive AI
Agentic AI is proactive rather than reactive, using AI apps — or agents — to comprehend the environment in which they are used without needing constant human control. Custom enterprise generative AI can deliver a more concentrated AI application than more generic programs such as ChatGPT and Midjourney with bespoke functionality for specific business requirements.
Radio technology developers have been working with AI, and some, including cloud-native automation and networking specialist Radio.Cloud, are already looking at what the latest iterations can offer the sector. “We are working heavily with LLMs, especially GPT-4o, and have thousands of prompts to create custom and unique AI voice tracks,” comments Andrew Scaglione, Radio.Cloud account director. “But we’re not doing any prompting with visual or audio input — it’s currently all text-based. ElevenLabs’ speech-to-speech AI voice-changer is a pretty cool tool that probably falls under the multimodal AI category. I see tools like that helping the industry for creative production, idea generation and more.”
How radio broadcasters use AI — specifically generative AI — now and how they will use new versions in the future varies wildly from country to country and organization to organization
Radio.Cloud has a traditional radio technology background and recently moved into AI for virtual automation and operations. By contrast, Veritone started working with AI on launching in 2014. According to the Managing Director of Media and Broadcast Paul Cramer, the company saw the potential of ANI (artificial narrow intelligence), the precursor to both generative and multimodal AI. “ANI are individual models that are very good at doing one particular thing,” he explains. “They’re trained on a large corpus of data and are domain-specific — such as transcription or translation. We have been doing that for almost a decade with our aiWARE operating system.”
He adds that despite the capabilities and potential of generative AI and beyond, people should look at practical use cases rather than what is referred to within Veritone as “Harry Potter use cases” because they sound almost like magic.
The crazy pace of progress
Zack Zalon, chief executive of Super Hi-Fi, observes that regardless of the technology approach, AI will have a massive impact in whatever industry it is used. “Specifically with radio, AI will reduce the time necessary for creating and managing world-class experiences,” he says. We strongly believe the result will be more exciting and dynamic content offerings to attract and retain the next generation of radio listeners.”
Like any technology, AI should be viewed in terms of what advantages it can bring and how it can move specific market areas forward, something with which Joe D’Angelo, senior vice president of global radio and digital audio at Xperi, agrees. “Radio stations are trying to proactively find that balance between using AI to streamline production and content creation while ensuring they preserve their status as the ‘trusted voice’ in the media landscape,” he says.
“The pace of innovation and the applications are astounding and, in some cases, very scary. I’ve experienced some new platforms with amazing capabilities to create long-form audio content and podcasts with little input and guidance. It’s not human creativity, but it’s pretty impressive and can certainly help with more routine content creation, such as weather and traffic reports.”
At voice-cloning software specialist Respeecher, Co-Founder and Chief Technology Officer Dmytro Bielievtsov says there is no “full-cycle AI on the radio” with intelligence that could be compared to ChatGPT. “Currently, GPT-4o does not realize it has a voice,” he says. “If you ask it to raise its voice a little, it will decline, saying it can’t do it since it’s texting you. OpenAI plans to release something more end-to-end as an alpha version soon, but we’ll have to see how well it works in practice. That said, the pace of progress in generative AI is crazy. There is some chance that we might be hitting a ceiling of how smart the current generation of AI can get, but we’re just about two years in, so clearly, not enough time has passed to make such a statement.”
Supplement rather than replace
How radio broadcasters use AI — specifically generative AI — now and how they will use new versions in the future varies wildly from country to country and organization to organization.
The BBC introduced 12 generative AI pilot schemes in October 2023 to maximize the value of existing content, create new audience experiences and make tasks quicker and easier. Rhodri Talfan Davies, director of BBC Nations, stated in an update published this February that a live radio sports commentary could be quickly converted to text for BBC Sport’s web pages as part of the scheme to reformat existing material.
Chris Ward, group operations director at Bauer Media Audio UK, says its network of commercial radio stations in the United Kingdom is “starting to see the impact of AI” by using the technology to supplement people rather than replace them and to speed up the production process. “Audio producers frequently encounter the need to make minor amendments to a recorded script,” he explains. “If, for example, the date in a recorded trail is incorrect, we can make the change instantly without needing to bring the voice artist back into the studio — with their consent, of course.” He adds that multimodal AI, in particular, is proving useful in training audio producers by suggesting specific changes to the mixing or mastering of a recording to enhance its sound.
These are still relatively early days for AI in radio, and presenters, journalists and engineers continue to work out what the technology can do for the medium. While phrases like “AI radio stations” and “AI presenters” are being bandied about, the smart thinking is that AI will remain an assistive technology to help support creativity and improve efficiency rather than replace the human element.
The author trained as a radio journalist and worked for British Forces Broadcasting Services Radio as a technical operator, producer and presenter before moving into magazine writing during the late 1980s. He recently returned to radio through his involvement in an online station where he lives on the south coast of England.
This article was taken from a RedTech special edition, “Radio Futures: Keeping an eye on AI” which you can read here.
More Bauer Media mentions
Bauer Media Audio agrees to acquire SharpStream
