Record sales expected for audio books21. June 2023
Record sales expected for audio books
San Francisco, 20.6. 2023
By 2030, audiobook sales are expected to exceed $35 billion annually, with AI playing an increasing role in production.
AI technology is already replacing human narrators in audio books, as seen in new offerings from Google Play and Apple Books. However, challenges remain in areas such as delivery, timing and pronunciation. For its digital narration, Apple combines advanced speech synthesis AI with the expertise of audio engineers, quality control specialists and linguists. Amazon’s Audible currently has a service for recording audiobooks, but it requires the audiobooks to be read by humans.
The decision to use AI voices in audio book production still depends on cost, time and listener preferences. Voice actors have different views on the impact of AI on their profession. Some believe that AI cannot match the nuance and emotion of human voices, while others see it as a useful tool for licensing their cloned voices.
With the advent of smartphones, the audio book industry has already seen a decade of significant growth. According to Wordsrated, audio book sales are projected to grow at an impressive 26.4% annually from 2022 to 2030. This makes it the fastest growing book format in the world, with the role of AI in audio book production expected to increase significantly.
Researchers from top universities warn about the problem of AI model collapse that occurs when generative AI models are trained on data created by other AI models. The process results in poor performance, more errors, and repetitive responses. Training models with machine-generated data rather than human-created data causes them to “forget” which could pose a problem as LLMs continue to contribute to the online language.
The breakdown of a model arises from the accumulation of errors during training, which skew the model’s understanding of reality. Recursive training exacerbates this problem and causes models to deviate further from the original data distribution.
In their study, the researchers simulated the effects of training generative models on their own data and observed complete changes in the data distribution within 50 generations. To maintain the quality of future generative models, the researchers emphasize the importance of training with human-made content and ensuring fair representation of minority groups in the datasets.
They suggest preserving the original human-made dataset, regularly including it in model training, and introducing new, clean human-made datasets.