A probabilistic framework for segment-based speech recognition

By James R. Glass. Computer Speech and Language, 2003

344 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/d6df5c972ed687aefe41637f3b0433c7dd44a0b6

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

In Signal Processing

Citations in the corpus

Phone-discriminating minimum classification error (p-MCE) training for phonetic recognition (2007, methodology): “In the phonetic recognizer’s evaluation, we merge the 48 monophones into 39 monophones according to the standard mapping described in [6, 7] and the confusion among the merged phones is not considered as errors.”
Conditional-Computation-Based Recurrent Neural Networks for Computationally Efficient Acoustic Modelling (2018, methodology): “In the past, the limitations imposed by the fixed-rate segmentation scheme inspired the so called segment-based ASR methods [9, 10, 11] that, in contrast to our approach, do not focus primarily on computational efficiency as they typically require many more decoding operations than frame-based syste”
The MIT Mobile Device Speaker Verification Corpus: Data Collection and Preliminary Experiments (2006, methodology): “Because our experiments utilize the segment-based SUMMIT speech recognition system [12], we have used segment-based acoustic feature vectors for speaker verification as well.”
Dynamic Graphical Models (2010, background): “DGMs generalize dynamic BNs (DBNs) [3], dynamic (or temporal) Markov random fields (DMRFs) (both of which themselves generalize hidden Markov models (HMMs) and hierarchical HMMs [6]), Boltzmann chains [27], sequential segmental models [7], dynamic conditional random fields (CRFs) x4”
Discriminative Training for Large-Vocabulary Speech Recognition Using Minimum Classification Error (2007, methodology): “SUMMIT’s standard real-time recognition configuration performs acoustic modelling using a set of 1388 context-dependent landmark models [68].”
Deep Belief Networks for phone recognition (2009, background): “This has led to promising results using segment-and landmark-based methods for phone recognition (e.g, [8] [9]).”
Unsupervised Speech Recognition via Segmental Empirical Output Distribution Matching (2018, methodology): “Another line of research uses a simpler segmentation method as an initialization, and jointly trains the segmenting and acoustic models for phonemes or words (Kamper et al., 2015; Glass, 2003; Siu et al., 2014; Lee and Glass, 2012).”
Supervised Sequence Labelling with Recurrent Neural Networks (2012, methodology): “, 2006) frequency-warped LPC cepstra were used as inputs, while Halberstadt and Glass tried a number of variations and combinations of MFCC, perceptual linear prediction (PLP) cepstral coefficients, energy and duration (Halberstadt, 1998; Glass, 2003).”
A Nonparametric Bayesian Approach to Acoustic Model Discovery (2012, methodology): “We follow the pre-segmentation method described in Glass (2003) to achieve the goal.”
Use of Differential Cepstra as Acoustic Features in Hidden Trajectory Modeling for Phonetic Recognition (2007, neutral): “The results are reported on the standard core testset of 192 utterances (24 speakers), the same setup as that described in [8].”
Fixed-dimensional acoustic embeddings of variable-length segments in low-resource settings (2013, background): “These include downsampling [13, 14, 15, 16], phonetic acoustic model-derived features [17, 18], and convolutional deep neural networks [19].”
On internal language representations in deep learning: an analysis of machine translation and speech recognition (2018, background): “Hybrid approaches (26)Examples include finite-state methods [245], discriminative training [71, 276], segment-based methods [124, 363], and a variety of language models [152].”
Unsupervised Lexicon Discovery from Acoustic Input (2015, methodology): “We follow the pre-segmentation method described in (Glass, 2003) to achieve the goal.”
Automatic alignment and error correction of human generated transcripts for long speech recordings (2006, methodology): “In the ﬁrst stage of our approach, we use the SUMMIT automatic speech recognition (ASR) system to automatically generate a transcript for a long audio ﬁle [7, 11].”
Dynamic Speech Models: Theory, Algorithms, and Applications (2006, methodology): “6% accuracy) obtained by using many heterogeneous classifiers as reported in [127] also.”
Learning Word Embeddings: Unsupervised Methods for Fixed-size Representations of Variable-length Speech Segments (2018, methodology): “[13] use mean pooling and temporal derivatives over subparts of each segment, duration and energy, while [14] warp segments into a fixed length.”
A Conversational Movie Search System Based on Conditional Random Fields (2012, methodology): “For the speech recognition of users’ utterances, we use the SUMMIT system [7], the acoustic models of which are trained with an English corpus unrelated to this domain.”
Unsupervised Pattern Discovery in Speech (2008, methodology): “In order to subdivide the audio stream into discrete segments of continuous speech, we use a basic phone recognizer to identify regions of silence in the signal [38].”