Cross-domain speaker recognition using domain adversarial siamese network with a domain discriminator

• Published in: Electronics letters Vol. 56; no. 14; pp. 737 - 739
• Main Authors: Chen, Zhigao, Miao, Xiaoxiao, Xiao, Runqiu, Wang, Wenchao
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 09.07.2020
• Subjects: adversarial‐learning mechanism; AISHELL‐Wake‐Up‐1 data set; automatic speaker recognition; background training; cross‐channel data; cross‐domain speaker recognition; deep neural networks; distribution mismatch; domain adversarial methods; domain adversarial siamese network; domain consistence; domain discriminator; domain distributions; domain influence; domain mismatch; domain‐invariant; evaluation data; Gaussian processes; learning (artificial intelligence); neural nets; NIST speaker recognition evaluation; speaker recognition; speaker‐discriminative; Speech and audio processing and translation; test data; unknown domain; unknown domain; background training; test data; domain adversarial siamese network; domain consistence; domain discriminator; evaluation data; cross-domain speaker recognition; domain-invariant; learning (artificial intelligence); deep neural networks; cross-channel data; domain influence; domain adversarial methods; automatic speaker recognition; NIST speaker recognition evaluation; speaker-discriminative; AISHELL-Wake-Up-1 data set; adversarial-learning mechanism; domain mismatch; Gaussian processes; domain distributions; neural nets; speaker recognition; distribution mismatch
• ISSN: 0013-5194; 1350-911X; 1350-911X
• DOI: 10.1049/el.2020.0673

With the widespread use of automatic speaker recognition in realistic world, it suffers a lot when there is a domain mismatch, including channel, language, distance etc. Recent research studies have introduced the adversarial-learning mechanism into deep neural networks to reduce the distribution mismatch between different domains. However, they only aligned the domain distributions between the background training and evaluation data. Few focused on the diverse distribution underlying the enrol and test data. In this Letter, the authors propose a domain adversarial siamese (DAS) network trying to eliminate the domain influence on speech representation. Specifically, they feed a network with speech pairs from the same speaker. Then a domain discriminator is introduced to capture the domain consistence or difference between pairs. Final embeddings become domain-invariant and more speaker-discriminative. A cross-channel data set is sort out from NIST speaker recognition evaluation and more experiments are conducted on AISHELL-Wake-Up-1 data set. Results show that DAS performs equally effective with typical domain adversarial methods, improving results at least $10\%$10% on energy efficiency rating. Furthermore, it is proved to be more valid for scenarios such as unbalanced data amount and unknown domain, achieving relatively $11\%$11% improvements.