We propose a novel framework for noise robust automatic speech recognition (ASR) based on cochlear implant-like spectrally reduced speech (SRS). Two experimental protocols (EPs) are proposed in order to clarify the advantage of using SRS for noise robust ASR. These two EPs assess the SRS in both the training and testing environments. Speech enhancement was used in one of two EPs to improve the quality of testing speech. In training, SRS is synthesized from original clean speech whereas in testing, SRS is synthesized directly from noisy speech or from enhanced speech signals. The synthesized SRS is recognized with the ASR systems trained on SRS signals, with the same synthesis parameters. Experiments show that the ASR results, in terms of word accuracy, calculated with ASR systems using SRS, are significantly improved compared to the baseline non-SRS ASR systems. We propose also a measure of the training and testing mismatch based on the Kullback-Leibler divergence. The numerical results show that using the SRS in ASR systems helps in reducing significantly the training and testing mismatch due to environmental noise. The training of the HMM-based ASR systems and the recognition tests were performed by using the HTK toolkit and the Aurora 2 speech database.
This letter considers high-rate block turbo codes (BTC) obtained by concatenation of two single-error-correcting Reed-Solomon (RS) constituent codes. Simulation results show that these codes perform within 1 dB of the theoretical limit for binary transmission over additive white Gaussian noise with a low-complexity decoder. A comparison with Bose-Chaudhuri-Hocquenghem BTCs of similar code rate reveals that RS BTCs have interesting advantages in terms of memory size and decoder complexity for veryhigh-data-rate decoding architectures. Index Terms-Block turbo codes (BTC), Reed-Solomon codes, product codes.
I. INTRODUCTIONSince their introduction in 1993 [1], turbo codes have received widespread interest within the digital communications community. The original concept of iterative softinput soft-output (SISO) decoding of concatenated convolutional codes has been extended to block turbo codes (BTC) [2] and to low-density parity-check (LDPC) codes [3]. Today, turbo codes have become a crucial industrial technology for error correction in digital transmission systems since they offer an excellent tradeoff between complexity and performance.This letter focuses on BTCs constructed from Reed-Solomon (RS) constituent codes. BTCs provide an interesting alternative to convolutional turbo codes and LDPC codes for applications requiring high code rates (R>0.8) and high-data-rate decoders. BTCs were originally introduced using binary Bose-Chaudhuri-Hocquenghem (BCH) constituent codes. BCH BTC is now an efficient and mature technology already in use in several proprietary satellite transmission systems, and has been recently adopted as an option in the IEEE 802.16 Standard for wireless Metropolitan Area Networks (MAN). Innovative architectures have been proposed that can achieve decoding speeds of several gigabits per second [4], [5]. An experimental demonstration of a forward error correction (FEC) for 10-Gb/s optical
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.