Deep neural models, particularly the long short-term memory recurrent neural network (LSTM-RNN) model, have shown great potential for language identification (LID). However, the use of phonetic information has been largely overlooked by most existing neural LID methods, although this information has been used very successfully in conventional phonetic LID systems. We present a phonetic temporal neural model for LID, which is an LSTM-RNN LID system that accepts phonetic features produced by a phone-discriminative DNN as the input, rather than raw acoustic features. This new model is similar to traditional phonetic LID methods, but the phonetic knowledge here is much richer: It is at the frame level and involves compacted information of all phones. Our experiments conducted on the Babel database and the AP16-OLR database demonstrate that the temporal phonetic neural approach is very effective, and significantly outperforms existing acoustic neural models. It also outperforms the conventional i-vector approach on short utterances and in noisy conditions.

Deep speaker embedding has demonstrated state-of-the-art performance in speaker recognition tasks. However, one potential issue with this approach is that the speaker vectors derived from deep embedding models tend to be non-Gaussian for each individual speaker, and non-homogeneous for distributions of different speakers. These irregular distributions can seriously impact speaker recognition performance, especially with the popular PLDA scoring method, which assumes homogeneous Gaussian distribution. In this paper, we argue that deep speaker vectors require deep normalization, and propose a deep normal-ization approach based on a novel discriminative normalization flow (DNF) model. We demonstrate the effectiveness of the proposed approach with experiments using the widely used SITW and CNCeleb corpora. In these experiments, the DNF-based normalization delivered substantial performance gains and also showed strong generalization capability in out-of-domain tests.

This paper proposes a solution for signal read-out in the MEMS cochlea sensors that have very small sensing capacitance and do not have differential sensing structures. The key challenge in such sensors is the significant signal degradation caused by the parasitic capacitance at the MEMS-CMOS interface. Therefore, a novel capacitive read-out circuit with parasitic-cancellation mechanism is developed; the equivalent input capacitance of the circuit is negative and can be adjusted to cancel the parasitic capacitance. Chip results prove that the use of parasitic-cancellation is able to increase the sensor sensitivity by 35 dB without consuming any extra power. In general, the circuit follows a low-degradation low-amplification approach which is more power-efficient than the traditional high-degradation high-amplification approach; it employs parasitic-cancellation to reduce the signal degradation and therefore a lower gain is required in the amplification stage. Besides, the chopper-stabilization technique is employed to effectively reduce the low-frequency circuit noise and DC offsets. As a result of these design considerations, the prototype chip demonstrates the capability of converting a 7.5 fF capacitance change of a 1-Volt-biased 0.5 pF capacitive sensor pair into a 0.745 V signal-conditioned output at the cost of only 165.2 μW power consumption.