William Hartmann
Senior scientist at Raytheon BBN Technologies. Research interests include automatic speech recognition and translation, domain adaptation, unsupervised learning, and reinforcement learning.
Projects
-
Seq2Seq ASR
Improving sequence-to-sequence models for low-resource, out-of-domain scenarios
-
Unsupervised Representation Learning
Applying unsupervised representation learning to speech and other signals
-
Code-Switching
Developing ASR models for code-switched langauges without data
-
MT for CTS
Machine translation for conversational speech
News
-
2022-10-28»
Training Autoregressive Speech Recognition Models with Limited in-domain Supervision
- new preprint submitted to IEEE ICASSP 2023
-
2022-05-09»
Combining Unsupervised and Text Augmented Semi-Supervised Learning for Low Resourced Autoregressive Speech Recognition
- Presented at IEEE ICASSP 2022
- 2021-10-07» Election to IEEE Speech and Language Technical Committee
- 2021-02-24» IEEE Senior Membership
Recent Posts
- 2023-01-30» Don Quixote and Reversing Expectations
- 2023-01-13» Training Autoregressive Speech Recognition Models with Limited In-Domain Supervision
- 2022-12-05» Sancho Panza Anticipates the Pale Blue Dot
- 2022-10-28» The Novels of Andy Weir
- 2022-09-19» Reconstruction vs. Contrastive Loss
Old Projects
-
IARPA MATERIAL
Cross-lingual information retrieval (CLIR) in both speech and text
-
Ship Identification
Adapting speaker identification technology to ship identification
-
IARPA BABEL
Program focused on keyword spotting in conversational speech for low-resourced languages
-
Acoustic Unit Discovery
Data-driven discovery of pronunciation units
-
Binary Mask Estimation
Estimating and applying binary masks for noise-robust ASR