The objective of this paper is to provide a summary of the state-of-the-art of automatic speech recognition (ASR) and its relevance to military applications. Until recently, speech recognition had its widest application in the development of vocoders for narrow-band speech communications. Presently, research in ASR has been accelerated for military tasks such as command and control, secure voice systems, surveillance of communication channels, and others. Research in voice control technology and digital narrow-band systems are of special interest. Much of the emphasis of today's military-supported research is to reduce to practice the current state of knowledge of ASR, as well as directing research in such a way as to have future military relevance. In coordination with the above-mentioned emphasis in military-supported research, this paper is divided into two major sections. The first section presents discussion of the state-of-the-art and problems in the various subareas of the ASR field. The second section presents a number of unsolved problems and techniques which need to be perfected before the solutions to a number of military applications of the ASR field are possible.

An assessment of the technology of automatic speech recognition for military applications

Traditional parametric coding of speech facilitates low rate but provides poor reconstruction quality because of the inadequacy of the model used. We describe how a WaveNet generative speech model can be used to generate high quality speech from the bit stream of a standard parametric coder operating at 2.4 kb/s. We compare this parametric coder with a waveform coder based on the same generative model and show that approximating the signal waveform incurs a large rate penalty. Our experiments confirm the high performance of the WaveNet based coder and show that the speech produced by the system is able to additionally perform implicit bandwidth extension and does not significantly impair recognition of the original speaker for the human listener, even when that speaker has not been used during the training of the generative model.

Wavenet based low rate speech coding

The voice mode of the Opus audio coder can compress wideband speech at bit rates ranging from 6 kb/s to 40 kb/s. However, Opus is at its core a waveform matching coder, and as the rate drops below 10 kb/s, quality degrades quickly. As the rate reduces even further, parametric coders tend to perform better than waveform coders. In this paper we propose a backward-compatible way of improving low bit rate Opus quality by re-synthesizing speech from the decoded parameters. We compare two different neural generative models, WaveNet and LPCNet. WaveNet is a powerful, high-complexity, and high-latency architecture that is not feasible for a practical system, yet provides a best known achievable quality with generative models. LPCNet is a low-complexity, low-latency RNN-based generative model, and practically implementable on mobile phones. We apply these systems with parameters from Opus coded at 6 kb/s as conditioning features for the generative models. A listening test shows that for the same 6 kb/s Opus bit stream, synthesized speech using LPCNet clearly outperforms the output of the standard Opus decoder. This opens up ways to improve the decoding quality of existing speech and audio waveform coders without breaking compatibility.

Improving Opus Low Bit Rate Quality with Neural Speech Synthesis

Rapid advances in machine-learning based generative modeling of speech make its use in speech coding attractive. However, the current performance of such models drops rapidly with noise contamination of the input, preventing use in practical applications. We present a new speech-coding scheme that is based on features that are robust to the distortions occurring in speech-coder input signals. To this purpose, we encourage the feature encoder to provide the same independent features for each of a set of linguistically equivalent signals, obtained by adding various noises to a common clean signal. The independent features, subjected to scalar quantization, are used as a conditioning vector sequence for WaveNet. Our experiments show that a 1.8 kb/s implementation of the resulting coder provides state-of-the-art performance for clean signals, and is additionally robust to noisy input.

Robust Low Rate Speech Coding Based on Cloned Networks and Wavenet

The standard first-order preemphasis approach used in linear prediction analysis results in an undesirable low-frequency boost in the synthesis spectrum. A solution is obtained by mismatching the preemphasis and postemphasis factors or by using a particular second-order preemphasis filter.

Spectral mismatch due to preemphasis in LPC analysis/Synthesis

This paper describes the real time implementation of a Linear Predictive Coding algorithm that has been developed over the past five years. The algorithm chosen for the analyzer is a modification of the Covariance Method introduced by B. S. Atal [1],[2] of Bell Labs. The system for pitch extraction uses a minimum distance function correlation technique. A dynamic programming algorithm [3] is used for pitch smoothing and correction of isolated pitch errors. The synthesizer uses a transversal filter. Considerable time has been devoted to optimizing the running time and integer scaling of the different algorithms for real time implementation on a 16 bit mini-computer.

T. Tremain

Papers

Linear predictive coding systems