Showing 1–2 of 2 results for author: Stiff, A
-
Spectral feature map** with mimic loss for robust speech recognition
Authors:
Deblin Bagchi,
Peter Plantinga,
Adam Stiff,
Eric Fosler-Lussier
Abstract:
For the task of speech enhancement, local learning objectives are agnostic to phonetic structures helpful for speech recognition. We propose to add a global criterion to ensure de-noised speech is useful for downstream tasks like ASR. We first train a spectral classifier on clean speech to predict senone labels. Then, the spectral classifier is joined with our speech enhancer as a noisy speech rec…
▽ More
For the task of speech enhancement, local learning objectives are agnostic to phonetic structures helpful for speech recognition. We propose to add a global criterion to ensure de-noised speech is useful for downstream tasks like ASR. We first train a spectral classifier on clean speech to predict senone labels. Then, the spectral classifier is joined with our speech enhancer as a noisy speech recognizer. This model is taught to imitate the output of the spectral classifier alone on clean speech. This \textit{mimic loss} is combined with the traditional local criterion to train the speech enhancer to produce de-noised speech. Feeding the de-noised speech to an off-the-shelf Kaldi training recipe for the CHiME-2 corpus shows significant improvements in WER.
△ Less
Submitted 26 March, 2018;
originally announced March 2018.
-
Substructure in the Andromeda Galaxy Globular Cluster System
Authors:
K. M. Perrett,
D. A. Stiff,
D. A. Hanes,
T. J. Bridges
Abstract:
In the most prominent current scenario of galaxy formation, galaxies form hierarchically through the merger of smaller systems. Such mergers could leave behind dynamical signatures which may linger long after the event. In particular, the globular cluster system (GCS) of a merging satellite galaxy may remain as a distinct sub-population within the GCS of a massive galaxy. Using the latest availa…
▽ More
In the most prominent current scenario of galaxy formation, galaxies form hierarchically through the merger of smaller systems. Such mergers could leave behind dynamical signatures which may linger long after the event. In particular, the globular cluster system (GCS) of a merging satellite galaxy may remain as a distinct sub-population within the GCS of a massive galaxy. Using the latest available globular cluster velocities and metallicities, we present the results of a search for grou** in the GCS of our nearest large spiral galaxy neighbor, M31. A modified friends-of-friends algorithm is used to identify a number of possible merger remnants in projected position, radial velocity and [Fe/H] parameter space. Numerical simulations are used to check that such merger remnants are indeed plausible over the timescales of interest. The identification of stellar streams associated with these groups is required in order to confirm that they represent merger remnants.
△ Less
Submitted 25 February, 2003;
originally announced February 2003.
