Blog(11)

• A More Accurate Internal Language Model Score Estimation for the Hybrid Autoregressive Transducer

  Language model (LM) adaptation in hybrid autoregressive transducer (HAT) is justified only when the transducer logits and the sum of speech and text logits in the label estimation sub-networks are approximately the same.

• A Model for Every User and Budget: Label-Free and Personalized Mixed-Precision Quantization

  Research in Automatic Speech Recognition (ASR) continues to show that larger models yield better results. But while state-of-the-art networks continue to grow with billions of parameters, the difficulty of deploying these models on device also increases.

• pMCT - Patched Multi-Condition Training

  The clashing of pans and pots as you cook and ask your voice assistant what you can use to replace eggs in the recipe. The excited, overlapping conversations as you ask which of Henry the VIIIs wives survived, trying to settle a bet.

View More

Focus Areas(0)

Research Achievements(30)

• FedNST: Federated Noisy Student Training for Automatic Speech Recognition

  AuthorHaaris Mehmood, Agnieszka Dobrowolska, Karthikeyan Saravanan, Mete Ozay

  PublishedAnnual Conference of the International Speech Communication Association(INTERSPEECH)

  Date2022-09-30

• pMCT: Patched Multi-Condition Training for Robust Speech Recognition

  AuthorPablo Peso Parada, Retiree, Karthikeyan Saravanan, Mete Ozay

  PublishedAnnual Conference of the International Speech Communication Association(INTERSPEECH)

  Date2022-09-19

• FFC-SE: Fast Fourier Convolution for Speech Enhancement

  AuthorIvan Shchekotov, Pavel Andreev, Aibek Alanov, Oleg Ivanov

  PublishedAnnual Conference of the International Speech Communication Association (INTERSPEECH)

  Date2022-09-18

View More

News(8)

• Samsung Research at SpeechBrain 2023 Online Summit - ISCA 2023 Interspeech Satellite Event

  The Samsung AI Center – Cambridge is co-organizing alongside other academic and industrial partners an official ISCA 2023 Interspeech Satellite Event taking the form of a one-day long “Online SpeechBrain Summit” on August 28th.

• [Blog] [INTERSPEECH 2022 Series #6] Multi-stage Progressive Compression of Conformer Transducer for On-device Speech Recognition

  Automatic Speech Recognition (ASR) systems on smart devices have traditionally relied on server based models. This involves sending audio data to the server and receiving text hypothesis once the server model completes decoding.

• [Blog] [INTERSPEECH 2022 Series #5] Prototypical Speaker-Interference Loss for Target Voice Separation Using Non-Parallel Audio Samples

  Deep learning based audio signal processing is increasingly becoming popular for solving the cocktail party problem [1-5]. A noisy signal, which is mixed by a clean and noise signal, is used as a pair with the clean signal for training a speech enhancement or source separation model [6-7].

View More

Others(0)