DECtalk Speech Synthesizer (1984)

From MSc Voice Technology
Revision as of 10:28, 18 October 2023 by 145.131.150.108 (talk)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Introduction[edit | edit source]

DECtalk is a mature text-to-speech synthesis technology developed and sold by Digital Equipment Corporation. Initially, it was developed by Digital's Assistive Technology Group (ATG) as an alternative to character cell terminals and for telephone applications.[1] DECtalk is a text-to-speech converter that provides any computer with a human-sounding voice. You can enter the text from a host computer or a terminal keyboard connected to the unit. You can hear the resulting speech through the built-in speaker, a headset, external speakers with an amplifier, or a telephone. You can also display the spoken output on a monitor screen or print it in hardcopy form.Today, DECtalk also provides a means for visually impaired individuals to access information. [2]

Historical Context[edit | edit source]

DECtalk's history reflects the evolution of speech synthesis technology. Advancements in computer power revolutionized voice input and output, enabling major corporations like Digital to offer telephone-based database access.[3] During the late 1970s and early 1980s, a surge of commercial products related to text-to-speech and speech synthesis flooded the market. One pivotal moment during this era occurred in 1983 with the introduction of the first commercial versions of DECtalk synthesizers, which represented a substantial advancement within the realm of speech synthesis technology. [1] It was widely used across various fields, particularly in the healthcare sector. Moreover, DECtalk Software, a product that has seen over 30 years of development, stands as a testament to natural-sounding, highly intelligible text-to-speech synthesis. It is accessible across various operating systems and signifies a momentous milestone in the historical trajectory of speech synthesis technology.[4]

Key Innovations[edit | edit source]

Dennis Klatt's pioneering work in formant-based synthesis, exemplified by DECtalk and later "KLSYN88," revolutionized text-to-speech systems. DECtalk's collection of standard voices, including the iconic "Perfect Paul," became famous. Notably used by Stephen Hawking, it served as a remarkable augmentative speaking device. DECtalk's innovation combined formant synthesis with articulatory methods, enhancing speech synthesis technology.[5]

DECtalk converts text from ASCII format to phonemic code using an internal dictionary and English pronunciation rules. Phonemic code is further processed by setting pitch and timbre. Finally, it generates speech with synthesizer control commands.[2]

The division in synthesis into two parts in DECTalk's BST system allows for easy configuration of synthesizers for various languages, including English, German, French, Spanish, and Japanese, by creating distinct sets of rules. The common feature is that both have used DECtalk as a formant synthesis system, providing dedicated processing modules which adapt their input according to the acoustic properties of the number of emotions.[6]

DECtalk has made significant innovations in the field of speech synthesis. Traditional concatenation techniques often led to choppy, unnatural speech with fixed intonation. DECtalk, on the other hand, employs a digital formant synthesizer that takes phonemic symbols as input, enabling better intonation control based on clause structure and more natural coarticulation effects. This approach significantly reduces storage requirements while allowing for the generation of multiple voices.[1]

Earlier methods relied on word databases or formant patterns, making it difficult to adjust pitch and duration within words.[7] Diphone synthesis, a popular technique, used actual speech segments for improved coarticulation but suffered from inconsistent pitch and intonation. DECtalk overcame these limitations by combining phonetic rules and vocal tract parameter modifications. One point of contention was how speech samples should be returned. Some argued for in-memory buffers, placing the audio system responsibility on the application. The chosen approach, however, supports standard workstation audio and includes an API call to switch the system into a speech-to-memory mode. This decision simplifies most applications' usage, ensuring a seamless experience for users without requiring extensive code rewrites with updates.[1]

High intelligibility scores in standard tests have recognized DECtalk's advancements in speech synthesis. Among text-to-speech systems, DECtalk is the preferred choice for visually impaired individuals. Its compact data footprint, at less than 1.5 megabytes, exemplifies efficient innovation in delivering high-quality, intelligible speech. DECtalk Software offers three unique modes to cater to various user needs: the speech-to-wave file mode, log-file mode, and speech-to-memory mode. Each mode has paired entry and exit calls. In the speech-to-wave file mode, the system records all speech samples into a wave audio file, which is closed upon exiting. This feature is particularly handy for slower Intel systems that can't handle real-time speech synthesis. The log-file mode captures the phonemic symbol output of the LTS thread, saving it to a file. The speech-to-memory mode is another standout innovation. When in this mode, an application uses a special API call to provide memory buffers to the text-to-speech system. The system then generates synthesized speech, returning it to the application via the same mechanism used for index marks. This mode offers flexibility and efficient memory handling.[1][8]

In formant synthesis field, Dennis Klatt introduced an advanced formant synthesizer, the Klatt Formant Synthesizer, which combined both cascade and parallel methods in 1980. It featured extra resonances and antiresonances for nasalized sounds, a sixth formant for high-frequency noise, and a bypass path for a flat transfer function. This complex excitation model, controlled by 39 parameters every 5 ms, improved sound modeling and was incorporated into DECtalk, enhancing its speech synthesis capabilities.[3] This is also a major innovation of DECtalk.

Impact[edit | edit source]

DECtalk has had a profound impact in the following fields, in addition to serving the function of General Text-to-Speech: Telecommunication Systems, Healthcare, Assistive Technologies, Robotics and Automation Systems, and Education.

First, DECtalk can be widely applied in automated phone systems, particularly in Interactive Voice Response (IVR) systems. IVR systems allow callers to interact with computers over the phone by using voice commands or touch-tone inputs.

Second, DECtalk plays an important role in healthcare filed. DECtalk's speech synthesis capabilities can be utilized to create assistive communication devices for individuals with speech or communication disabilities. These devices help patients who may have conditions like ALS (Amyotrophic Lateral Sclerosis) or other speech-affecting disorders to express themselves by converting text input into speech. Stephen Hawking, who suffers from ALS, is a user of DECtalk, and DECtalk helped him make sounds.[9] Besides that, DECtalk can be integrated into medical transcription software. DECtalk makes it possible to provide full digital speech output for all medical files and terms.[10] In addition to the speech impaired people, DECtalk plays a crucial role in creating screen readers and text-to-speech tools for visually impaired individuals. These tools convert on-screen text into spoken words, enabling users to access digital content, including websites and documents.

In addition to the above two important points, DECtalk also provides the robot with the ability to speak. In other words, it is used in the robot's voice. In the field of education, DECtalk generates audio as a material to help educators and students understand something, especially to help students with visual impairments.[11]

Future research[edit | edit source]

For DECtalk, which had just come out at that time, there were still many problems. First of all, although many advanced speech technologies were utilized by it, the sounds it generated at the time were still far from human speech.[1] Secondly, its eight built-in voices cannot completely imitate the speaker's native pronunciation, nor can they express the speaker's emotions. The pitch variations of the speech it generates are not diverse. Additionally at the time, DECtalk lacked American English pronunciation, although this was implemented in the early 2000s. At present, DECtalk is gradually being replaced by other more advanced speech synthesis applications with deep learning AI as the main technology.[12] I think that although DECtalk plays a very important role in the history of speech synthesis, at present, deep learning AI trained through large-scale data can help us achieve more realistic speech and capture more details.

In addition, DECtalk has been widely used in the healthcare field, and I think that in the future, this field will be the most important development direction of DECtalk. DECtalk should realize Text to speech neural systems and achieve good quality voice output through deep learning AI.[13]


LLM Review[edit | edit source]

I asked ChatGPT: "You are my speech sound professor. Please evaluate my Wiki Page according to the following dimensions: language coherence, logic, and structure. Please point out the advantages and disadvantages." Then i sent my first draft to it.

Chatgpt pointed out my shortcomings in the following two points:

Language Coherence: While the overall language coherence is good, there are a few sentences that could be rephrased to enhance clarity and flow. For example, in the "Historical Context" section, the sentence "Moreover, DECtalk Software, a product that has seen over 30 years of development, stands as a testament to natural-sounding, highly intelligible text-to-speech synthesis" could be made more concise and clear.

Logic: While the page logically progresses from one section to another, some transitions between sections could be smoother. For example, you might consider adding transition sentences to help readers move from the historical context to key innovations seamlessly.

Regarding these two points, I modified some sentences myself and continued to seek suggestions from ChatGPT. The sentences in my impact part use DECtalk as the subject too much, which seems very repetitive, and rarely use connecting words. As a non-native English speaker who is still learning and improving my English, I did have a lot of problems with the logic and coherence of my first draft, and I accepted most of ChatGPT's suggestions.

Contributors[edit | edit source]

Ding Shenghuan

References[edit | edit source]

  1. 1.0 1.1 1.2 1.3 1.4 1.5 Hallahan W I. DECtalk software: Text-to-speech technology and implementation[J]. Digital Technical Journal, 1995, 7(4): 5-19.
  2. 2.0 2.1 Digital Equipment Corporation. DECtalk DTC01 Owner’s Manual. 1st ed., December 1983; 2nd ed., May 1984. Copyright ® 1983, 1984 by Digital Equipment Corporation. Printed in U.S.A. 1983: 1-9.
  3. 3.0 3.1 Heggtveit P O. An overview of text-to-speech synthesis[J]. TELEKTRONIKK, 2003, 99(2): 30-44.
  4. Fruchterman J. Developing Information Technology to Meet Social Needs (Innovations Case Narrative: Benetech)[J]. Innovations: Technology, Governance, Globalization, 2008, 3(3): 83-99.
  5. Story B H. History of speech synthesis[J]. Teoksessa The Routledge Handbook of Phonetics, toimittaneet William F. Katz & Peter F. Assmann, 2019: 9-33.
  6. Balyan A, Agrawal S S, Dev A. Speech synthesis: a review[J]. International Journal of Engineering Research & Technology (IJERT), 2013, 2(6): 57-75.
  7. D. Klatt, “Review of Text-to-Speech Conversion for English,” Journal of the Acoustical Society of America, vol. 82, no. 3 (1987): 737–793.
  8. Olofsson A. Phonemic awareness and the use of computer speech in reading remediation: Theoretical background[J]. Fonetiks, 1988, 1(5): 15-27.
  9. Kewley-Port D, M Nearey T. Speech synthesizer produced voices for disabled, including Stephen Hawking[J]. The Journal of the Acoustical Society of America, 2020, 148(1): R1-R2.
  10. Grams R R, Smillov A, Li B. Field-testing the new DECtalk PC system for medical applications[J]. Journal of medical systems, 1992, 16: 31-38.
  11. Leong C K. Effects of on-line reading and simultaneous DECtalk auding in helping below-average and poor readers comprehend and summarize text[J]. Learning Disability Quarterly, 1995, 18(2): 101-116.
  12. Ning Y, He S, Wu Z, et al. A review of deep learning based speech synthesis[J]. Applied Sciences, 2019, 9(19): 4050.
  13. Bolimera R, Kishore K T. Artificial Intelligence and IoT-based Healthcare System using TTS Assistive Technology for Voice Disordered People[J]. NeuroQuantology, 2022, 20(10): 3649-3656.