Towards Robust Distant-Talking Automatic Speech Recognition in Reverberant Environments

Sehr, Armin; Kellermann, Walter

doi:10.1007/978-3-540-70602-1_18

Cited by 10 publications

(6 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The melspec convolution (1) provides a very good approximation of the exact relationship between x mel (k) and s mel (k − m) [2] and is used as a constraint in the following.…”

Section: Review Of the Remos Conceptmentioning

confidence: 99%

“…Since an estimatex r,ij,mel (k) of the reverberant component x r,mel (k) can be readily obtained from the clean-speech estimateš s ij,mel (k − m) of previous iterations (see [2]) according to 2…”

Section: Remos In the Logmelspec Domainmentioning

confidence: 99%

“…To model the reverberant logmelspec Feature Vector Sequence (FVS) x(k) without the limitation of the conditional independence assumption, REMOS [1,2] uses a combination of an HMM network N λ modeling the clean speech and an RVM η as depicted in Fig. 1.…”

Section: Review Of the Remos Conceptmentioning

confidence: 99%

“…The reduced ASR performance is mainly caused by the dispersive effect of reverberation on the feature representation of speech used for ASR [2]. Since the Room Impulse Response (RIR) describing the acoustic path between speaker and microphone extends over several feature frames, the reverberation causes the speech features to be smeared along the time axis so that an effect similar to intersymbol interference known from radio communications is observed.…”

Section: Introductionmentioning

confidence: 98%

“…The subscript ij of, e.g.,x r,ij,mel (k),š ij,mel (k − m) indicates that these quantities depend on current state j and previous state i (see[2]). …”

mentioning

confidence: 99%

See 4 more Smart Citations

Model-based dereverberation in the logmelspec domain for robust distant-talking speech recognition

Sehr

Maas

Kellermann

2010

2010 IEEE International Conference on Acoustics, Speech and Signal Processing

Self Cite

View full text Add to dashboard Cite

The REMOS (REverberation MOdeling for Speech recognition) concept for reverberation-robust distant-talking speech recognition, introduced in [1] for melspectral features, is extended in this contribution to logarithmic melspectral (logmelspec) features. Based on a combined acoustic model consisting of a hidden Markov model network and a reverberation model, REMOS determines clean-speech and reverberation estimates during recognition by an inner optimization operation. A reformulation of this inner optimization problem for logmelspec features, allowing an efficient solution by nonlinear optimization algorithms, is derived in this paper so that an efficient implementation of REMOS for logmelspec features becomes possible. Connected digit recognition experiments show that the proposed REMOS implementation significantly outperforms reverberantlytrained HMMs in highly reverberant environments.

show abstract

“…The melspec convolution (1) provides a very good approximation of the exact relationship between x mel (k) and s mel (k − m) [2] and is used as a constraint in the following.…”

Section: Review Of the Remos Conceptmentioning

confidence: 99%

Section: Remos In the Logmelspec Domainmentioning

confidence: 99%

Section: Review Of the Remos Conceptmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 98%

“…The subscript ij of, e.g.,x r,ij,mel (k),š ij,mel (k − m) indicates that these quantities depend on current state j and previous state i (see[2]). …”

mentioning

confidence: 99%

See 3 more Smart Citations