Spaces:

amphion

/

PicoAudio

like 21

Running on Zero

Hi Hugging Face Team:

I am writing to apply for the GPU Community Grant on behalf of our new work PicoAudio (https://arxiv.org/abs/2407.02869v2) in Amphion, which focuses on developing controllable audio generation models.

PicoAudio enables millisecond-level timestamp and frequency control, transforming free-form text into structured text via LLM, which is then utilized in generating audio through a finely-grained controlled diffusion process. It represents a promising exploration in controlled audio generation frameworks, transitioning from intriguing generation to practical control. This innovative approach has the potential to revolutionize the way we generate and manipulate audio content, offering new tools and capabilities for researchers and developers in various domains. Moreover, we aim to identify what is more practical and valuable for both users and the industry. We aspire to develop an audio generation framework that is not only intriguing but also effective. The temporal control performance of PicoAudio may become a crucial element in achieving this goal.

I have shared my work on Hugging Face, where it is accessible to the broader research community. However, the performance of the current implementation on CPU-based systems is significantly hindered by slow processing times. This limitation not only affects the user experience but also restricts the full potential of the project's impact. To address this, the integration of GPU technology is essential, as it would drastically improve processing speeds and enable real-time results.

I am eager to potentially partner with your organization and deeply appreciate your consideration of my application.

Thank you for your attention.