STCON System for the CHiME-8 Challenge
Authors:
Anton Mitrofanov,
Tatiana Prisyach,
Tatiana Timofeeva,
Sergei Novoselov,
Maxim Korenevsky,
Yuri Khokhlov,
Artem Akulov,
Alexander Anikin,
Roman Khalili,
Iurii Lezhenin,
Aleksandr Melnikov,
Dmitriy Miroshnichenko,
Nikita Mamaev,
Ilya Odegov,
Olga Rudnitskaya,
Aleksei Romanenko
Abstract:
This paper describes the STCON system for the CHiME-8 Challenge Task 1 (DASR) aimed at distant automatic speech transcription and diarization with multiple recording devices. Our main attention was paid to carefully trained and tuned diarization pipeline and speaker counting. This allowed to significantly reduce diarization error rate (DER) and obtain more reliable segments for speech separation a…
▽ More
This paper describes the STCON system for the CHiME-8 Challenge Task 1 (DASR) aimed at distant automatic speech transcription and diarization with multiple recording devices. Our main attention was paid to carefully trained and tuned diarization pipeline and speaker counting. This allowed to significantly reduce diarization error rate (DER) and obtain more reliable segments for speech separation and recognition. To improve source separation, we designed a Guided Target speaker Extraction (G-TSE) model and used it in conjunction with the traditional Guided Source Separation (GSS) method. To train various parts of our pipeline, we investigated several data augmentation and generation techniques, which helped us to improve the overall system quality.
△ Less
Submitted 17 October, 2024;
originally announced October 2024.
End-to-End Autoencoder for Drill String Acoustic Communications
Authors:
Iurii Lezhenin,
Aleksandr Sidnev,
Vladimir Tsygan,
Igor Malyshev
Abstract:
Drill string communications are important for drilling efficiency and safety. The design of a low latency drill string communication system with high throughput and reliability remains an open challenge. In this paper a deep learning autoencoder (AE) based end-to-end communication system, where transmitter and receiver implemented as feed forward neural networks, is proposed for acousticdrill stri…
▽ More
Drill string communications are important for drilling efficiency and safety. The design of a low latency drill string communication system with high throughput and reliability remains an open challenge. In this paper a deep learning autoencoder (AE) based end-to-end communication system, where transmitter and receiver implemented as feed forward neural networks, is proposed for acousticdrill string communications. Simulation shows that the AE system is able to outperform a baseline non-contiguous OFDM system in terms of BER and PAPR, operating with lower latency.
△ Less
Submitted 6 May, 2024;
originally announced May 2024.
