Where Do We Want To Go Next?
Among the biggest technology stories of 2022, it's impossible not to reference OpenAI. This artificial intelligence (AI) research company from San Francisco, CA, founded in December 2015, has focused on artificial general intelligence (AGI) applied to augmenting and outperform human skills. OpenAI cleverly made its efforts accessible to the general public as a way to simultaneously promote AI and improve it with actual human interaction — which in turn helps to reveal boundaries and usage restriction needs.
Among OpenAI's most interesting efforts is a trained model for human interaction through text input called ChatGPT. Its proven uses span from helping us to debug code to answering questions, to writing essays about any topic and in any given suggested style. While in research preview, usage of ChatGPT remains free (chat.openai.com).
Another notable technology trend in 2022, from an early fascination with voice recognition and virtual assistants in always-connected smart speakers, the industry is now transitioning to AI/ML on-device (edge AI) — particularly for hearables. As I intended to write about the possibilities of AI/ML in audio applications, I asked ChatGPT about this topic.
Very quickly, ChatGPT identified the uses of AI/ML in speech recognition for transcription, in voice assistants and text-to-speech systems, and also for music classification (metadata indexing and recommendation engines), and music generation — curiously one of the earlier uses to create copyright-free music beds and even full original soundtracks in homemade videos. An interesting and somewhat surprising suggestion was the use of AI/ML for audio compression, dynamically preserving the quality of the original audio signal for storing or transmission. Which could mean an adaptive audio codec, but also choosing codecs according to conditions and requirements. I assume ChatGPT was referencing actual applications.
Finally, and as expected, ChatGPT mentions audio enhancement applications. "AI can be used to improve the quality of audio signals, such as by removing noise or by increasing the volume of quiet sounds."
In this issue of audioXpress, we address the fast-evolving world of MEMS microphones and voice pickup sensors, increasingly combined in true wireless earbuds, headphones, headsets, smartphones, and other personal systems. And one of the current trends is to add on-device AI for audio enhancement. To run machine learning tasks directly on edge devices, we can take advantage of the latest ARM Cortex-M and RISC-V architectures, as well as deep neural network processors and advanced processors specialized for both machine learning and audio processing such as those from Syntiant, BrainChip, FemtoSense Synaptics, memBrain, and many others.
Much like OpenAI is showing us unheard of possibilities, in 2023 we can expect to see audio applications and performance levels we have never imagined.
AI/ML shows infinite possibilities for audio product design when combined with modeling and simulation, which will enable us to reach the most efficient audio amplifiers with the lowest distortion, or near-perfect speaker diaphragms for a given set of parameters.
AI/ML tasks for audio DSP are actually very limited in scope compared to other tasks that are being required from current available engines. Models for audio applications are light enough to be loaded into miniature, ultra-low power designs. Of course, we also want to be able to use full-featured voice assistants, which require connectivity and are the reason why we will continue to connect wearable devices to smartphones (the most powerful connected computers we ever carried in our pockets).
The topic of connectivity is another major one for 2023. One of the unpredictable effects of the pandemic was that network operators had an unexpected reprieve that allowed them to accelerate the transition and coverage deployment of 5G. While we were all confined at home, the priority was home fixed broadband and Wi-Fi — an area that saw important progress with 6/6E and will soon evolve to Wi-Fi 7. As we traveled out of the home and upgraded to the latest smartphone models, we are now discovering the possibilities of being connected to 5G networks. A completely new world for audio streaming that is the new normal in 2023, while the industry is already working toward a 6G future.