Domain Adaptation Using Factorized Hidden Layer for Robust Automatic Speech Recognition

By Khe Chai Sim, Arun Narayanan, Ananya Misra, Anshuman Tripathi, Golan Pundak, Tara N. Sainath, Parisa Haghani, Bo Li, Michiel Bacchiani. 2018

37 citations recorded by the source corpus.

Domain robustness is a challenging problem for automatic speech recognition (ASR). In this paper, we consider speech data collected for different applications as separate domains and investigate the robustness of acoustic models trained on multi-domain data on unseen domains. Specifically, we use Factor-ized Hidden Layer (FHL) as a compact low-rank representation to adapt a multi-domain ASR system to unseen domains. Experimental results on two unseen domains show that FHL is a more effective adaptation method compared to selectively fine-tuning part of the network, without dramatically increasing the model parameters. Furthermore, we found that using singular value decomposition to initialize the low-rank bases of an FHL model leads to a faster convergence and improved performance.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Jinyu Li
2. Khe Chai Sim
3. Tara N. Sainath
4. Arun Narayanan
5. Trevor Strohman
6. Petr Zadrazil
7. Michiel Bacchiani
8. Bo Li

In Representation & Self-Supervised Learning

1. Khe Chai Sim
2. Michiel Bacchiani

In Speech & Audio

1. Khe Chai Sim
2. Trevor Strohman
3. Michiel Bacchiani
4. Arun Narayanan
5. Tara N. Sainath
6. Petr Zadrazil
7. Golan Pundak
8. Parisa Haghani

In Signal Processing

1. Khe Chai Sim
2. Jinyu Li
3. Trevor Strohman
4. Arun Narayanan
5. Tara N. Sainath
6. Michiel Bacchiani
7. Petr Zadrazil
8. Bo Li

Citations in the corpus

• Toward Domain-Invariant Speech Recognition via Large Scale Training (2018, background): “It is also possible to introduce domain specific parameters, as in [36], to avoid performance degradation when fine-tuning to a certain subset.”
• Personalization of End-to-End Speech Recognition on Mobile Devices for Named Entities (2019, neutral): “This is consistent with the results reported in [12], where robust domain adaptation can be achieved by fine-tuning the entire model with millions of parameters on a small amount of adaptation data.”
• An Investigation Into On-device Personalization of End-to-end Automatic Speech Recognition Models (2019, result): “This is inline with the results reported for domain adaptation [29].”