#interspeech2020

DATASET RELEASE: "CC100", the CommonCrawl dataset of 2.5TB of clean unsupervised text from 100 languages (used to train XLM-R) is now publicly available. You can find below the Data: data.statmt.org/cc-100/ Script: bit.ly/3oC6aXy By @VishravC et al.

Better ways of using unlabeled data for #TeacherStudentLearning in #E2E & hybrid #ASR. Zhang et al & Sapru et al describe #SemiSupervised learning for ASR at #INTERSPEECH2020. bit.ly/3efpsNX bit.ly/3oswnrw Ian McLoughlin Amazon Science tomorrow at thu-1-2

Subword regularisation of #E2E #ASR to handle unseen words and improve performance. Egor Lakomkin et al propose modifications to E2E training. #interspeech2020 bit.ly/37OHu8w Amazon Science Tomorrow Thu-1-2

Wilfried Michel et al propose to use text data while training #acousticmodels for #ASR. Their technique local fusion improves performance and is almost as fast as vanilla training. #INTERSPEECH2020 bit.ly/3kNVqD2 AppTek Tomorrow Thu-1-2

#Convolutional layers vs #Transformer layers. ContextNet by Wei Han et al at #INTERSPEECH2020 proposes how to use convolutional neural layers to achieve on par performance with Transformers for #E2E ASR with fewer parameters. bit.ly/3kCzqv3 Google AI Tomorrow Thu-1-2

Simpler architecture for cache-based neural #LanguageModel can outperform complex neural cache models overall as well as on rare words. Ke li et al describe their novel solution at #INTERSPEECH2020. bit.ly/3jJ2kIM JHU CLSP Tomorrow Thu-1-3

Using BERT is better than uni-directional Transformer as a teacher for training Sequence-to-Sequence ASR. Hayato Fatumi et al introduce a training method for applying #BERT #LanguageModel to seq2seq #ASR. bit.ly/2JgVJbR Hirofumi Inaguma Tomorrow Thu-1-3

Improved pointer mechanism for #contextualising #LanguageModels from meta-data. Da-Rong Liu et al introduce a hybrid pointer mechanism improving ASR rescoring results. #ASR #INTERSPEECH2020. bit.ly/3oCnGuY Facebook AI Tomorrow Thu-1-3

Augment #LanguageModel target data by performing domain transfer on text. Atsunori Ogawa et al from NTT detail the augmentation schemes which improves #ASR performance. bit.ly/2HI693P #INTERSPEECH2020 #NTT Tomorrow Thu-3-8

i've listened to a few interviews in this awesome series. the only complaint i have about this interview series is its name. these are more like "super"humans of AI rather than humans of AI. cc.gatech.edu/~parikh/humans…

A few of these AI-generated thought experiments are amazingly good (others are silly or trivial, but that's ok). It's remarkable what AI systems like GPT3 can do now. HT: Eliezer Yudkowsky ⏹️ cc: Christopher Manning Andrew Ng Steve Jurvetson Max Tegmark #AI

👋 Hi, we're Coqui :) Open speech tech 🐸💬 is our mission 🚀 💚 Join the community 👩‍💻 Use the tech 💪 Share the power Leave a 🌟 on Github for new models and releases! github.com/coqui-ai

Finetuning Pretrained Transformers into RNNs Successfully converts a pretrained transformer into its efficient linear-complexity recurrent counterpart with a learned feature map to improve the efficiency while retaining the accuracy. arxiv.org/abs/2103.13076

Join our webinar Building operational AI/ML with MLOps on thu April 15 with 🔸 Kaisa Salakka, Director, Research Labs at Unity Technologies, 🔸 Jukka Remes , PhD, Lead AI Solutions Architect 🔸 Teppo Kuisma, VP, Business Development ➡️ learn.silo.ai/webinar-mlops #ML #MLOps #AIforpeople

The MIT Computational Psycholinguistics Lab seeks to fill an open postdoc position for an MIT-IBM Watson AI Lab supported multi-PI project in low-resource language learning, bridging NLP, machine learning, linguistics, and cognitive science. Please spread the word! careers.peopleclick.com/careerscp/clie…

Webinar this week's Thursday: Building operational AI/ML w/ MLOps 🔸 Kaisa Salakka, Director, Research Labs at Unity Technologies, 🔸 Jukka Remes , PhD, Lead AI Solutions Architect 🔸 Teppo Kuisma, VP, Business Development ➡️ Sign up learn.silo.ai/webinar-mlops #ML #MLOps #AIforpeople

🔥🔥 Coqui community model zoo is Open 🔥🔥 ✅ Speech-to-Text models for 36 languages ✅ Models for on-device (TensorFlow Lite) ✅ Easily customize with your own Language Model Kudos to the model creators and community members who made this possible 🎉 coqui.ai/models

Anyone with Python skills and a masters looking for a 13 month full time job? Knowing java, German or working with language data a plus but not a must. Can be done from home office. Applications close tomorrow (14 April) so be quick!

AI-driven sensor fusion is seen as the enabling technology behind autonomous vehicles & vessels. Our Senior AI Scientist Jesus Carabano shares his experience from working with sensor fusion in the automotive & maritime industries: silo.ai/real-world-app… #sensorfusion #ML #AI