A series of virtual meetings to set the path forward
Dates of the webinars
The Kaldi speech recognition toolkit, born in Johns Hopkins University (2009) and debuted at the Prague ICASSP (2011), is undergoing a metamorphosis.
Carefully designed, scalable deep learning algorithms, implemented well before frameworks such as TensorFlow, PyTorch, MxNet or Caffe were widely adopted, have been key to Kaldi’s good performance and widespread adoption in academia and industry. Equally important are the many recipes
that encapsulate the knowledge and tradecraft of Kaldi experts in service of the Kaldi user/developer community.
Meanwhile, deep learning frameworks have evolved tremendously in the last 10 years. Both algorithmic innovation in and scalable deployment of AI systems is increasingly able to take advantage of such frameworks, significantly accelerating progress.
We are therefore undertaking the first major revision of Kaldi, aimed at integrating it closely with standard deep learning infrastructure used in research and practice. We are calling on the Kaldi researcher-, developer- and user-community to come together once again to build this shared resource.
Note: For now, we only host archive of the past events. Get alerted about future meetings by watching our repository on github
The first community meeting will focus on the research community, both academic and non-academic, and engage past, current and future Kaldi users and contributors. Participants will share their thoughts on how to make Kaldi easier to teach, learn and modify for both recurring and bespoke research projects. Read more
The second community meeting will focus on applications built with Kaldi components, both commercial and non-commercial, and engage past, current and future developers of Kaldi-based solutions and products. Participants will share their thoughts on how to make Kaldi easier to customize, scale up, deploy, and maintain in real-time and off-line use cases. Read more
The third community meeting will engage creators of deep learning frameworks and computing infrastructures, to understand trends in framework development and hardware acceleration, insights into existing or upcoming features of platforms that may fulfill Kaldi needs, and possibilities for interoperability or interchangeability of ASR tools and models between frameworks. Read more