HI-MIA: A Far-Field Text-Dependent Speaker Verification Database and the Baselines

Abstract

This paper presents a far-field text-dependent speaker verification database named HI-MIA. We aim to meet the data requirement for far-field microphone array based speaker verification since most of the publicly available databases are single channel close-talking and text-independent. The database contains recordings of 340 people in rooms designed for the far-field scenario. Recordings are captured by multiple microphone arrays located in different directions and distance to the speaker and a high-fidelity close-talking microphone. Besides, we propose a set of end-to-end neural network based baseline systems that adopt single-channel data for training. Moreover, we propose a testing background aware enrollment augmentation strategy to further enhance the performance. Results show that the fusion systems could achieve 3.29% EER in the far-field enrollment far field testing task and 4.02% EER in the close-talking enrollment and far-field testing task.

Meta-Learning for Cross-Channel Speaker Verification

“…Furthermore, to validate the effect of MSEN on unseen channels, we design the generalized crosschannel (GCC) evaluation. The results on the HI-MIA corpus [18] demonstrate that MSEN drastically reduces the adverse impact of channel mismatch on recognition results, and significantly outperforms the state-of-the-art methods.…”

Unit Selection Synthesis Based Data Augmentation for Fixed Phrase Speaker Verification

“…In this work, we use the AISHELL2 1 [15] as text-independent training data, AISHELL-wakeup 2 [11] as text-dependent training data, and AISHELL-2019B-eval dataset 3 [11] as text-dependent test set.…”

“…The AISHELL-2019B-eval contains recordings of 86 speakers with Chinese wake-up word "ni hao, mi ya".The room setting and recording devices are the same as that of AISHELL-wakeup. Utterances of the last 44 people are selected as the test set since they are more challenging [11]. This corpus has two tasks: close-talking enrollment task (utterances from the close-talking mic are used for enrollment) and far-field enrollment task (utterances from one 16-channel circular microphone array which is 1m away from the speaker are used for enrollment).…”

“…Additive angular margin loss [17,18] with m=0.2 and s=32.0 is used as the projection layer since it has shown better performance than other losses. For the enrollment or test utterances that have 16 channels from a 6-channel circular microphone array, we adopt the strategy of speaker embedding level averaging as do in [11]. AISHELL2 is firstly used to train the text-independent x-vector model.…”

“…In practice, to achieve the state-of-the-art performance, text-dependent SV requires the same set of text to be spoken during the training stage and the test stage. [9,10,11]. When there is no or limited training data with the designated phrase to build a text-dependent SV system, those data augmentation approaches couldn't help to get promising performance.…”