I-vector Transformation Using Conditional Generative Adversarial Networks for Short Utterance Speaker Verification

Abstract

I-vector based text-independent speaker verification (SV) systems often have poor performance with short utterances, as the biased phonetic distribution in a short utterance makes the extracted i-vector unreliable. This paper proposes an i-vector compensation method using a generative adversarial network (GAN), where its generator network is trained to generate a compensated i-vector from a short-utterance i-vector and its discriminator network is trained to determine whether an i-vector is generated by the generator or the one extracted from a long utterance. Additionally, we assign two other learning tasks to the GAN to stabilize its training and to make the generated ivector more speaker-specific. Speaker verification experiments on the NIST SRE 2008 "10sec-10sec" condition show that our method reduced the equal error rate by 11.3% from the conventional i-vector and PLDA system.

Improved RawNet with Feature Map Scaling for Text-independent Speaker Verification using Raw Waveforms

“…al. [19], where the discriminative power of a DNN's high-dimensional intermediate representation can differ significantly with small perturbations. In addition, we assume that applying an additive FRM combined with a multiplicative FRM will lead to further improvements.…”

“…In addition, by applying an FRM through adding, we expect to provide small perturbations that lead to increased discriminative power. This is inspired by a previous study [19] that showed analyzing small alterations in high-dimensional space can drastically change discriminative power. By hypothesizing that these two approaches function in a complementary manner, we also propose to apply both approaches in sequence.…”

The Second DIHARD Challenge: System Description for USC-SAIL Team

“…Motivated by the success of conditional generative adversarial network (CGAN) in speech enhancement [9] and i-vector transformation in short-utterance speaker verification [10], we propose to use CGAN on x-vector embeddings to compensate for additive noise in the speaker diarization framework. The approach is to train CGAN using both clean and noisy x-vectors, which can generate denoised x-vectors from noisy input xvectors.…”

“…We use Wasserstein GAN [11] model in the CGAN framework. Furthermore, similar to [10], we incorporate multi-task training of G, where the generator network is integrated with another network Gsup for speaker prediction. The first section G is optimized to simul-taneously reduce the generator loss, mean square error (MSE) loss and cross-entropy (CE) loss.…”

Training Multi-task Adversarial Network for Extracting Noise-robust Speaker Embedding

“…Its great success in image processing has inspired people to consider whether it can also be applied into the field of speech processing. In the paper [11], Zhang et al attempted to use conditional GAN to solve the impact of performance degradation caused by the variable-duration of utterances. Ding et al [12] proposed a multi-tasking GAN framework to extract the more distinctive speaker representation.…”