DARPA Speech Understanding Research

From MSc Voice Technology
Revision as of 19:35, 19 September 2023 by Youyang Cai (talk | contribs)
Jump to navigation Jump to search

Introduction

DARPA which stands for Defense Advanced Research Projects Agency, is a research agency of the United States Department of Defense that responsible for the development of emerging technologies, among which in the 70s was speech recognition, for use by the military. As the DARPA's then-director George Heilemeir declared, "Get computers to read Morse code in the presence of other code and noise, get computers to identify/detect key words in a stream of speech, [...] make a real contribution to command and control, and; do a good thing in sonar"[1]. Consequently, the project received funding from the U.S. Department of Defense, particularly the Navy, given its potential military applications.

Given the military context, the project was subject to specific strict requirements. Notably, it needed to recognize multiple speakers simultaneously and achieve real-time speech recognition with no delays. As a result, the project's research objectives were defined as follows:

  • Accepting connected speech
  • Recognizing speech from multiple cooperative speakers
  • Accepting 1,000 words
  • Yielding only <10% semantic errors
  • Achieving real-time understanding

This is the reason why it is understanding in the first place, and not recognition: to attain successful understanding what was said or what was intented to be said rathern than simple recognizing some words taken out of context.

Historical Context

While there had been partially successful attempts to understand discrete speech (see, for example, Bell's Audrey), there were virtually no systems capable of comprehending continuous speech at the time, except for Raj Reddy's recognition system, which was primarily used for issuing chess commands[2]. Furthermore, previous methods were limited to vocabularies of no more than 200 words (e.g., IBM's 16-word "Shoebox"). In contrast, DARPA's SUR project aimed to achieve speech recognition with a vocabulary of at least 1,000 words[3]. As a result, the project's goals significantly surpassed the capabilities of existing state-of-the-art solutions.

Project Progess

Research groups were established at Carnegie Mellon University (CMU), SRI International, MIT's Lincoln Laboratory, Systems Development Corporation (SDC), and Bolt, Beranek, and Newman (BNN). CMU's research efforts resulted in the development of two systems, HARPY and HEARSAY-II, while BNN was responsible for creating Hear What I Mean (HWIM)[4]. These three systems developed under the project represent different technical solutions, and, as a result, different performances.

HARPY

Harpy demonstrated the ability to recognize speech using a vocabulary of 1,011 words with reasonable accuracy of 90%[3]. A significant contribution from the Harpy system was the introduction of a graph search concept, where language was represented as a connected network derived from lexical representations of words, incorporating syntactical production rules and also the rules for word boundaries.

In this system, the input speech underwent parametric analysis, followed after by segmentation. The segmented speech sequence was then subjected to phone template matching using the Itakura distance metric. The graph search, based on a beam algorithm, compiled, hypothesized, pruned, and verified the recognized sequence of words or sounds that best satisfied knowledge constraints, achieving the highest matching score which was defined as the smallest distance to the reference patterns. Notably, the Harpy system was among the first to leverage a finite state network to reduce computational load and efficiently identify the closest matching string[5], which began to be widely used in speech recognition problems especially after the invent of Hidden Markov Models.

HEARSAY-II

At the heart of this system, Hearsay-II put what are known as symbolic problem solvers, also referred to as knowledge sources. The need for multiple knowledge sources stemmed from the diverse transformations applied by speakers when creating acoustic signals and the corresponding inverse transformations required by listeners for interpretation[6].

These knowledge sources communicated through a blackboard architecture, which served as a repository for data, partial results, and finished conclusions. This design allowed each knowledge source to know where to retrieve information from the blackboard and where to post partial conclusions. The range of knowledge sources included phonetic, phonemic, lexical, syntactic, semantic, prosodic, discursive, and even psychological aspects. Each of these sources independently proposed improved word string guesses for the given speech signal by 'writing' them on the virtual blackboard. Other knowledge sources could then build upon these suggestions.

Using this technique, Hearsay achieved recognition of 1,011 words in continuous speech from multiple speakers with limited syntax, achieving an accuracy rate of approximately 90%[7]. However, a notable limitation of this system was the time spent deciding which knowledge source to utilize next, which detracted from real-time speech processing — contrary to one of the project's primary goals.

HWIM

Similar to Hearsay-II, the Hear What I Mean system was also knowledge-based, but it employed a more explicit scheduling approach based on human problem-solving methods. The process began with the identification of the most certain words, which served as 'islands of certainty,' and then leveraged these to iteratively expand both bottom-up and top-down identification processes.[8]

Unlike using a central mechanism as in Hearsay-II, in this system, knowledge sources would communicate with each other and share processed data[7]. To score phoneme and word hypotheses, Bayesian probabilities were utilized. Furthermore, the system represented syntax using an Augmented Transition Network, presenting a greater challenge at the syntax level compared to previous systems. As a result, this system was able to accept 1,097 words, however, yielding more than 50% semantic error[9] and thus not meeting the goals of the project.

General overview

DARPA's program, despite achieving the set goals, did not result in a speech understanding system for daily or even military use. The resulting systems were too restricted in terms of syntax, yielded semantic errors and required large computational resources. Performances of the different systems were also difficult to compare because of the different vocabularies and domains employed. The Hearsay-II and HARPY results, however, are comparable, as the two systems were tested on the same tasks using the same test data, with HARPY's performance dominating Hearsay-II's in both accuracy and computation speed. In general, HARPY was the only system clearly to meet and exceed the DARPA specifications.

Key Innovations

Despite the fact that these systems were not ready for use there and then, the research did discover and elucidate much new information about speech, and developed new architectural insights, particularly the blackboard architecture that has since been used in other AI systems.

The notable outcome of this program was a system capable of accurately identifying 90%[10] of human-generated utterances from a vocabulary of more than 1,000 words, a significantly larger vocabulary size than previous approaches. Most suited to the goals of the project, Harpy achieved this by accessing word meanings from a database and determining sentence structure using its fast 'beam search' technology, a novel application of this approach. Furthermore, when Harpy encountered speech it couldn't understand, it responded with an 'I don't know what you said, please repeat' message, a feature reminiscent of today's voice assistants[11].

Impact

As a result of this project, it became evident that machines could be trained to comprehend not just individual words and numbers, but entire sentences. Consequently, subsequent research shifted its focus from discrete to continuous speech and headed towards making the modeling of syntactic rules and speech recognition overall more complex and sophisticated.

Future research

In this article, our primary focus has been on the relatively successful (Harpy) or semi-successful (Hearsay-II, HWIM) outcomes of the project. However, it's important to note that some institutions, such as SDC, also participated in the project but did not achieve significant results. Therefore, it is worthwhile to shift the attention to their approaches and explore the reasons behind their lack of success.

LLM review

I

References

  1. Gaon, A. H. (2021). The future of copyright in the age of artificial intelligence. Edward Elgar Publishing.
  2. Reddy, D. R., Erman, L. D., Fennell, R. D., Lowerre, B. T., & Neely, R. B. (1974). The Hearsay speech understanding system. The Journal of the Acoustical Society of America, 55(2_Supplement), 409-409.
  3. 3.0 3.1 Furui, S. (2005). 50 years of progress in speech and speaker recognition research. ECTI Transactions on Computer and Information Technology (ECTI-CIT), 1(2), 64-74.
  4. Klatt, D. H. (1977). Review of the ARPA speech understanding project. The Journal of the Acoustical Society of America, 62(6), 1345-1366.
  5. Juang, B. & Rabiner, Lawrence. (2005). Automatic Speech Recognition - A Brief History of the Technology Development.
  6. Erman, L. D., Hayes-Roth, F., Lesser, V. R., & Reddy, D. R. (1980). The Hearsay-II speech-understanding system: Integrating knowledge to resolve uncertainty. ACM Computing Surveys (CSUR), 12(2), 213-253.
  7. 7.0 7.1 Erman, L. D., Hayes‐Roth, F., Lesser, V. R., & Reddy, R. (1976). The Hearsay‐II speech understanding system. The Journal of the Acoustical Society of America, 60(S1), S11-S11.
  8. Schwartz, R., Barry, C., Chow, Y. L., Deft, A., Feng, M. W., Kimball, O., ... & Vandegrift, J. (1989). The BBN BYBLOS continuous speech recognition system. In Speech and Natural Language: Proceedings of a Workshop Held at Philadelphia, Pennsylvania, February 21-23, 1989.
  9. Lowerre, B.T. (1976). The HARPY speech recognition system.
  10. Thorndyke, P. W., & Reddy, R. (1989, August). High-Impact Future Research Directions for Artificial Intelligence. In IJCAI (p. 1675).
  11. A brief history of voice assistants. The Verge [URL: https://www.theverge.com/ad/17855294/a-brief-history-of-voice-assistants]

Group members

  • Igor Marchenko
  • Wangyiyao Zhou
  • Yanpei (Page) Ouyang
  • Youyang Cai
  • Yi Lei