The INTERSPEECH 2020 Far-Field Speaker Verification Challenge

Abstract

The INTERSPEECH 2020 Far-Field Speaker Verification Challenge (FFSVC 2020) addresses three different research problems under well-defined conditions: far-field text-dependent speaker verification from single microphone array, far-field textindependent speaker verification from single microphone array, and far-field text-dependent speaker verification from distributed microphone arrays. All three tasks pose a cross-channel challenge to the participants. To simulate the real-life scenario, the enrollment utterances are recorded from close-talk cellphone, while the test utterances are recorded from the far-field microphone arrays. In this paper, we describe the database, the challenge, and the baseline system, which is based on a ResNetbased deep speaker network with cosine similarity scoring. For a given utterance, the speaker embeddings of different channels are equally averaged as the final embedding. The baseline system achieves minDCFs of 0.62, 0.66, and 0.64 and EERs of 6.27%, 6.55%, and 7.18% for task 1, task 2, and task 3, respectively.

Task-specific Optimization of Virtual Channel Linear Prediction-based Speech Dereverberation Front-End for Far-Field Speaker Verification

“…Three different ASV benchmarks Voices Obscured in Complex Environmental Settings (VOiCES) Challenge 2019 [4], Speakers in the Wild (SITW) [39], and Far-field Speaker Verification Challenge (FFSVC) 2020 [5] are utilized to validate the proposed approach.…”

“…The FFSVC 2020 dataset [5] contains Chinese Mandarin utterances recorded with a cellular phone and six four-channel circular microphone arrays at a distance. The speaker enrollment and tests are conducted using the cellular phone and one of the microphone array recordings, respectively.…”

“…The speaker enrollment and tests are conducted using the cellular phone and one of the microphone array recordings, respectively. We adopt "Task 2" [5], which is a text-independent ASV task using a single microphone array, and only use the first channel recordings to compute the ASV scores. The cellular phone recordings are downsampled to 16 kHz.…”

“…To achieve ASV performances competitive to those reported in [5], [57], [58] on the FFSVC 2020 benchmark, we additionally fine-tuned the pretrained DSE model using the "train" portion of the FFSVC 2020 dataset. We concatenated the recordings and obtained 4,920 utterances from 120 speakers.…”

“…Moreover, to prevent the potential degeneration of the front-end in non-far-field speaking conditions, we propose a distortion regularization method within the TSO framework by mimicking the response of the WPE algorithm to nearfield speech recordings. Experiments were conducted on two far-field ASV benchmarks [4], [5] and one in-the-wild ASV benchmark [39] to further examine the generalizability of the front-ends to unconstrained conditions. The main contributions of this article are summarized as follows:…”