Adaptive Multimodal Fusion by Uncertainty Compensation With Application to Audiovisual Speech Recognition

Abstract

Abstract-While the accuracy of feature measurements heavily depends on changing environmental conditions, studying the consequences of this fact in pattern recognition tasks has received relatively little attention to date. In this paper, we explicitly take feature measurement uncertainty into account and show how multimodal classification and learning rules should be adjusted to compensate for its effects. Our approach is particularly fruitful in multimodal fusion scenarios, such as audiovisual speech recognition, where multiple streams of complementary time-evolving features are integrated. For such applications, provided that the measurement noise uncertainty for each feature stream can be estimated, the proposed framework leads to highly adaptive multimodal fusion rules which are easy and efficient to implement. Our technique is widely applicable and can be transparently integrated with either synchronous or asynchronous multimodal sequence integration architectures. We further show that multimodal fusion methods relying on stream weights can naturally emerge from our scheme under certain assumptions; this connection provides valuable insights into the adaptivity properties of our multimodal uncertainty compensation approach. We show how these ideas can be practically applied for audiovisual speech recognition. In this context, we propose improved techniques for person-independent visual feature extraction and uncertainty estimation with active appearance models, and also discuss how enhanced audio features along with their uncertainty estimates can be effectively computed. We demonstrate the efficacy of our approach in audiovisual speech recognition experiments on the CUAVE database using either synchronous or asynchronous multimodal integration models.Index Terms-Active appearance models (AAMs), audiovisual automatic speech recognition (AV-ASR), multimodal fusion, uncertainty compensation.

Geometrical-based lip-reading using template probabilistic multi-dimension dynamic time warping

“…For visual features the approach used histogram-based descriptors around twelve lip landmarks determined using an AAM fitting technique and the classification involved multiple kernel learning and SVM. Similar results were reported by Papandreou et al [68] who achieved a best recognition rate of 83% in speaker independent experiments when using AAM visual features obtained from the entire lower face with six shape and six texture coefficients and when using HMM for classification.…”

Robust audiovisual speech recognition using noise-adaptive linear discriminant analysis

“…As already noted by [31], in coupled HMM decoding, stream weight adaptation and uncertainty compensation by UD both provide significant advantages in isolation, but using uncertainty compensation in addition to optimized stream weighting provides only small benefits. This finding was replicated in our experiments.…”

“…Uncertainty Decoding (denoted by GDU in the following tables) was used successfully for audiovisual speech recognition in [31]. In conjunction with uncertainty propagation techniques and stream weight optimization, however, the respective performance gains of UD become small.…”

Visual Speech Recognition Using Optical Flow and Support Vector Machines

“…These approaches were all proposed by Matthews et al 7 to extract visual features. Continuing their work on visual feature extraction, Papandreou et al 32 focused on multimodal fusion scenarios, using audiovisual speech recognition as an example. They demonstrated that their visemic AAM (based on digits 0-9) with six texture coefficients outperforms their PCA-based technique with 18 texture coefficients, achieving a word accuracy rate of 83% and 71%, respectively.…”