Deepdub, a pioneer in AI-driven dubbing solutions, has made its proprietary technology AudioSample available as an open-source, solution-oriented tool for the community to bring powerful new audio projects to life.
AudioSample will provide the tools necessary to efficiently handle audio data from loading and processing to integration with machine learning models.
Its abilities range from tackling massive audio datasets to seeking efficient ways to integrate audio processing into workflows.
Initially created as a tool to robustly and efficiently fuel Deepdub’s audio processing and machine learning advancements, particularly audio data for its PyTorch data loader, it quickly evolved – making an impact across a number of important areas, including the creation of a fast and efficient .wav file reader.
“AudioSample was born out of a frustration that many audio engineers and researchers share: the lack of a powerful yet efficient package to process large audio files,” said Nir Krakowski, CTO and Co-Founder of Deepdub. “Four years ago, we faced significant challenges dealing with extensive audio datasets. Existing tools fell short in terms of speed and functionality, particularly when integrated into our machine learning workflows. This motivated us to develop AudioSample, a solution tailored to our needs and now available for the community.”
Among its core features and innovations is its Lazy Loading capability, which ensures that audio data is loaded only when needed, which significantly optimizes memory usage and processing time – particularly for applications dealing with large audio files but only requiring specific segments.
Initially designed to handle .wav files, AudioSample has evolved and now supports a wide range of audio file formats after its integration with PyAV (powered by libffmpeg). This development allows users to work with multiple audio formats seamlessly, expanding the usability of AudioSample beyond its original scope.
For audio industry developers, researchers and engineers, AudioSample integrates with both PyTorch and NumPy, allowing them to incorporate it into existing pipelines without disrupting their workflow, facilitating more efficient and streamlined audio processing.
Engineered for projects requiring real-time audio data processing or extensive data manipulation, AudioSample ensures rapid processing times even when handling extensive audio files, delivering speed without compromising efficiency.
Since its inception, AudioSample has added new features to meet the evolving needs of the audio industry.