Zobrazeno 1 - 10
of 128
pro vyhledávání: '"Mouchtaris, Athanasios"'
Fine-tuning large language models (LLMs) has achieved remarkable performance across various natural language processing tasks, yet it demands more and more memory as model sizes keep growing. To address this issue, the recently proposed Memory-effici
Externí odkaz:
http://arxiv.org/abs/2406.18060
Autor:
Shakiah, Suhaila M., Swaminathan, Rupak Vignesh, Nguyen, Hieu Duy, Chinta, Raviteja, Afzal, Tariq, Susanj, Nathan, Mouchtaris, Athanasios, Strimel, Grant P., Rastrow, Ariya
Publikováno v:
IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar, 2023, pp. 100-107
Machine learning model weights and activations are represented in full-precision during training. This leads to performance degradation in runtime when deployed on neural network accelerator (NNA) chips, which leverage highly parallelized fixed-point
Externí odkaz:
http://arxiv.org/abs/2305.07778
Autor:
Fu, Xuandi, Sathyendra, Kanthashree Mysore, Gandhe, Ankur, Liu, Jing, Strimel, Grant P., McGowan, Ross, Mouchtaris, Athanasios
Attention-based contextual biasing approaches have shown significant improvements in the recognition of generic and/or personal rare-words in End-to-End Automatic Speech Recognition (E2E ASR) systems like neural transducers. These approaches employ c
Externí odkaz:
http://arxiv.org/abs/2305.05271
Autor:
Strimel, Grant P., Xie, Yi, King, Brian, Radfar, Martin, Rastrow, Ariya, Mouchtaris, Athanasios
Streaming speech recognition architectures are employed for low-latency, real-time applications. Such architectures are often characterized by their causality. Causal architectures emit tokens at each frame, relying only on current and past signal, w
Externí odkaz:
http://arxiv.org/abs/2305.04159
Autor:
Sahai, Saumya Y., Liu, Jing, Muniyappa, Thejaswi, Sathyendra, Kanthashree M., Alexandridis, Anastasios, Strimel, Grant P., McGowan, Ross, Rastrow, Ariya, Chang, Feng-Ju, Mouchtaris, Athanasios, Kunzmann, Siegfried
We present dual-attention neural biasing, an architecture designed to boost Wake Words (WW) recognition and improve inference time latency on speech recognition tasks. This architecture enables a dynamic switch for its runtime compute paths by exploi
Externí odkaz:
http://arxiv.org/abs/2304.01905
Autor:
Chang, Feng-Ju, Alexandridis, Anastasios, Swaminathan, Rupak Vignesh, Radfar, Martin, Mallidi, Harish, Omologo, Maurizio, Mouchtaris, Athanasios, King, Brian, Maas, Roland
To achieve robust far-field automatic speech recognition (ASR), existing techniques typically employ an acoustic front end (AFE) cascaded with a neural transducer (NT) ASR model. The AFE output, however, could be unreliable, as the beamforming output
Externí odkaz:
http://arxiv.org/abs/2303.00692
Autor:
Zhen, Kai, Radfar, Martin, Nguyen, Hieu Duy, Strimel, Grant P., Susanj, Nathan, Mouchtaris, Athanasios
For on-device automatic speech recognition (ASR), quantization aware training (QAT) is ubiquitous to achieve the trade-off between model predictive performance and efficiency. Among existing QAT methods, one major drawback is that the quantization ce
Externí odkaz:
http://arxiv.org/abs/2210.09188
Autor:
Radfar, Martin, Barnwal, Rohit, Swaminathan, Rupak Vignesh, Chang, Feng-Ju, Strimel, Grant P., Susanj, Nathan, Mouchtaris, Athanasios
The recurrent neural network transducer (RNN-T) is a prominent streaming end-to-end (E2E) ASR technology. In RNN-T, the acoustic encoder commonly consists of stacks of LSTMs. Very recently, as an alternative to LSTM layers, the Conformer architecture
Externí odkaz:
http://arxiv.org/abs/2209.14868
Autor:
Xie, Yi, Macoskey, Jonathan, Radfar, Martin, Chang, Feng-Ju, King, Brian, Rastrow, Ariya, Mouchtaris, Athanasios, Strimel, Grant P.
We present a streaming, Transformer-based end-to-end automatic speech recognition (ASR) architecture which achieves efficient neural inference through compute cost amortization. Our architecture creates sparse computation pathways dynamically at infe
Externí odkaz:
http://arxiv.org/abs/2207.02393
Autor:
Zhen, Kai, Nguyen, Hieu Duy, Chinta, Raviteja, Susanj, Nathan, Mouchtaris, Athanasios, Afzal, Tariq, Rastrow, Ariya
We present a novel sub-8-bit quantization-aware training (S8BQAT) scheme for 8-bit neural network accelerators. Our method is inspired from Lloyd-Max compression theory with practical adaptations for a feasible computational overhead during training.
Externí odkaz:
http://arxiv.org/abs/2206.15408