Revisiting SincNet: An Evaluation of Feature and Network Hyperparameters for Speaker Recognition

Oneață, Dan; Georgescu, Lucian Puiu; Cucu, Horia; Burileanu, Dragoş; Burileanu, Corneliu

doi:10.23919/eusipco47968.2020.9287794

Cited by 5 publications

(3 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A reduction of the Sinc filter widths to 31 samples (2 ms) gave a further improvement, especially when the stride was concurrently changed to 2 samples from 1 sample. This is consistent with the observations of previous work that smaller filter widths in the Sinc layer are superior in the context of speaker recognition [23]. While this seems counter-intuitive given that auditory filter impulse responses at lower centre frequencies are of duration well over 10 ms, the reduced frequency resolution due to the apparent truncation does not seem to harm the performance.…”

Section: Single-task Architecturessupporting

confidence: 91%

“…The first variation involves replacing the CNN layer at the input with bandpass constrained, but tunable, filters motivated by the traditionally used mel filterbanks emulating the low-level auditory processing [21]. SincNet has been applied in frame-level speaker identification where its hyperparameters have been found to be critical, although sometimes counter-intuitive, to achieved task accuracy [23]. Next, we try to exploit the linguistic association between phrase boundaries and prominent words with multi-task learning.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep Learning For Prominence Detection In Children's Read Speech

Vaidya¹,

Sabu²,

Rao³

2021

Preprint

View full text Add to dashboard Cite

The detection of perceived prominence in speech has attracted approaches ranging from the design of linguistic knowledge-based acoustic features to the automatic feature learning from suprasegmental attributes such as pitch and intensity contours. We present here, in contrast, a system that operates directly on segmented speech waveforms to learn features relevant to prominent word detection for children's oral fluency assessment. The chosen CRNN (convolutional recurrent neural network) framework, incorporating both word-level features and sequence information, is found to benefit from the perceptually motivated SincNet filters as the first convolutional layer. We further explore the benefits of the linguistic association between the prosodic events of phrase boundary and prominence with different multi-task architectures. Matching the previously reported performance on the same dataset of a random forest ensemble predictor trained on carefully chosen hand-crafted acoustic features, we evaluate further the possibly complementary information from hand-crafted acoustic and pre-trained lexical features.

show abstract

Section: Single-task Architecturessupporting

confidence: 91%

Section: Introductionmentioning

confidence: 99%

Deep Learning For Prominence Detection In Children's Read Speech

Vaidya¹,

Sabu²,

Rao³

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The model architectures were manually designed to reduce the model complexity for implementing lightweight SV [42]- [44]. In the work of [42], the original trunk of the SincNet [45] was replaced by a lightweight trunk with 2.8 million parameters for reducing the model complexity. Lee et al [44] designed a hyperbolic ResNet for lightweight application.…”

Section: Related Workmentioning

confidence: 99%

Few-Shot Speaker Identification Using Lightweight Prototypical Network With Feature Grouping and Interaction

Chen

Cao

et al. 2023

IEEE Trans. Multimedia

View full text Add to dashboard Cite

Although many efforts have been made on decreasing the model complexity for speaker verification, it is still challenging to deploy speaker verification systems with satisfactory result on low-resource terminals. We design a transformation module that performs feature partition and fusion to implement lightweight speaker verification. The transformation module consists of multiple simple but effective operations, such as convolution, pooling, mean, concatenation, normalization, and element-wise summation. It works in a plug-and-play way, and can be easily implanted into a wide variety of models to reduce the model complexity while maintaining the model error. First, the input feature is split into several low-dimensional feature subsets for decreasing the model complexity. Then, each feature subset is updated by fusing it with the inter-feature-subsets correlational information to enhance its representational capability. Finally, the updated feature subsets are independently fed into the block (one or several layers) of the model for further processing. The features that are output from current block of the model are processed according to the steps above before they are fed into the next block of the model. Experimental data are selected from two public speech corpora (namely VoxCeleb1 and VoxCeleb2). Results show that implanting the transformation module into three models (namely AMCRN, ResNet34, and ECAPA-TDNN) for speaker verification slightly increases the model error and significantly decreases the model complexity. Our proposed method outperforms baseline methods on the whole in memory requirement and computational complexity with lower equal error rate. It also generalizes well across truncated segments with various lengths.

show abstract