Mechanical synthesis: Difference between revisions

From MSc Voice Technology
Jump to navigation Jump to search
No edit summary
No edit summary
Line 64: Line 64:


This revised version offers a clearer organization of ideas and enhanced readability.</blockquote>Suggestion 1 was dismissed because it contains contradictory information and because I considered the timeline to reflect my intentions well and clearly enough. I have minimally revised some longer sentences; and I have completely dismissed the revised paragraph, as I think the style is excessively pompous and more convoluted than before.
This revised version offers a clearer organization of ideas and enhanced readability.</blockquote>Suggestion 1 was dismissed because it contains contradictory information and because I considered the timeline to reflect my intentions well and clearly enough. I have minimally revised some longer sentences; and I have completely dismissed the revised paragraph, as I think the style is excessively pompous and more convoluted than before.
=== Contributing student ===
Maria Tepei


== References ==
== References ==

Revision as of 23:20, 18 October 2023

Introduction

Speech synthesis refers to the artificial reproduction of human-like speech.[1] The very early attempts at speech synthesis can be traced back to as early as the 18th century[2], reflecting the undeniable significance of speech as a primary mode of human communication. Over time, technological advancements have greatly enhanced the accuracy, quality, and naturalness of synthesized speech; see for example projects such as OLGA[3], which integrated audiovisual information and facial animation to boost speech intelligibility. In recent years, this scientific field has witnessed significant advancements driven by extensive research, expanding computational capabilities, the accessibility of speech data, and substantial financial investments in this domain[4].

Historical Context

The broad context of the very first attempts at mechanical speech synthesis of the 1700s is the Enlightenment as a cultural and historical era, together with its fundamental ideas. This historical era encompassed ideas centred around human well-being, knowledge acquisition through rational thinking and sensory evidence, as well as freedom and scientific advancement. This brought about an overall growing recognition of the human intellect, with a strong emphasis on the scientific approach and principles of reductionism[5].

The 17th century had already brought about a clear understanding of the head an neck anatomy, as evidenced by the illustrations of the sagital sections found in anatomy handbooks and scientific articles[6][7]. However, there was still no clear understanding of how these components were responsible for speech production, as well as where the sound of voice exactly comes from[8].

One of the most detailed accounts of speech production and proposed speech synthesis mechanisms is given by Marin Mersenne[7] in 1636, where he provides clear instructions on constructing a set of organ pipes capable of 'articulating vowels, consonants, syllables, and complete utterances'. His work, along with later contributions form Denis Dodart [9] and Antoine Ferrein [10], were already correctly assuming the critical role of a tube's shape and manipulation thereof in reproducing speech sounds.

As of the 1750s, the theoretical framework for mechanisms underlying speech production revolved around the idea that a pair of lungs generated an aerodynamic flow of air, which passes through the vocal folds, causing them to vibrate under tension and create sound. This sound is subsequently modulated into speech through a tube-shaped vocal tract. In other words, at this point there was already no doubt regarding the analogy between the vocal tract and a 'special musical instrument', with the scientific focus shifting towards how this unique vocal instrument can be effectively played to produce speech[8]. At the same time, Leonhard Euler had already brought his major contribution to the field of physics regarding the understanding of sound waves, speculating from early on that it might be possible to build some kind of music instrument that would generate sounds which come close to human speech.

All together, these contributions from different fields in the context of a science-driven Enlightenment era mark a crucial juncture in the early understanding of speech production and potential approaches to speech synthesis.

Christian Kratzenstein: Generating vowels with acoustic resonators

Christian Gottlieb Kratzenstein is known in phonetics for being one of the earliest scientists to openly share his endeavours in speech synthesis. His achievement to approximate the synthesis of the vowels /a, e, i, o, u/ through resonating pipes is notable, given the fact that the understanding of acoustic resonance was still a work in progress. Through a process of trial and error[11], Kratzenstein's work reveals that the complexities of speech can be explained in physical and physiological terms, while highlighting that the principles behind speech synthesis are somewhat similar to those governing musical instruments.

Born in 1723 in Wemingerode, Germany, Kratzenstein studied natural sciences at the University of Halle, where he obtained his doctorate in 1746. He then became a professor of mechanics at the St. Petersburg University (1748-1753) and later at the University of Copenhagen (from 1753 until his death in 1795).

In 1779, Leonhard Euler proposes a competition at the Academy of Sciences in St. Petersburg, asking whether it might be possible to construct a set of pipes which would perfectly immitate the vowels /a, e, i, o, u/ [12]. Kratzenstein submits the work he had been gathering on this topic for over five years[11] and wins the competition by describing the nature of the differences between the five vowels in question, as well as implementing their 'synthesis' (rather: approximation) through resonating pipes. Building on prior information known at the time about speech production, Kratzenstein provides precise measurements of the jaw movements and lip openings as cues for differentiating the acoustic features of vowels; then he proceeds to build five flute-like acoustic resonators made of wood and metal, which he likely shaped through trial and error to create approximations of the five vowel sounds when blown with a free reed. A detailed account of this approach to mechanical speech synthesis was first published in Latin, shortly followed by a French translation[11].

Key Innovations. Impact

The era of mechanical 'talking machines' in general, and Kratzenstein's approach in particular, laid the fundamental groundwork for the development and practical applications of phonetic theory. These approaches were instrumental in advancing our understanding of the fundamental principle that resonances created by specific configurations of air cavities accentuate the spectral components of a sound source[12]. Even in these early attempts at mechanical speech synthesis, it became evident that a complex interplay of elements, including flow of air (like the one from the lungs), vibratory sound source (such as the larynx), and sound shaping through a tube, was necessary to achieve speech-like sound. These insights influenced the further developments of phonetic theory primarily by proving that the intuitions that had existed so far were valuable, correct, and practically reproducable.

Furthermore, the source-filter theory of speech production, explicitly articulated and introduced in the 1960s, built upon principles that had been implicitly recognized in Kratzenstein's and von Kempelen's endeavours of mechanical speech synthesis. The need for a source of sound, as well as the acoustic shaping of sound through a tube were present even in these initial attempts, a historical continuity which underscores the relevance of early mechanical efforts in shaping our theoretical understanding of speech production.

Last, but not least, one of the enduring challenges in phonetics, the many-to-one relationship between vocal tract geometry and acoustics, originated in the early experiments such the one of Christian Kratzenstein, and is still -- though maybe not to the same extent -- a topic of debate in the field. Overall, these early mechanical attempts, often perceived at that time as merely a source of amusement and enetertainment, encapsulate the early views on theories of speech production, as some of the challenges encountered continue to be a subject of contemporary research, demonstrating the lasting impact and relevance of the insights gained from these early mechanical speech synthesis endeavours[8].

LLM Review

ChatGPT was critically used for the purpose of improving the readability, coherence, grammaticality, and information flow in this article. Two different prompts were employed by the author to this end:

  • Evaluate the language use, coherency, and flow of information in the following paragraph.
  • Assess the accessibility and quality of explanations in the following text, considering it is intended to have a neutral, yet engaging tone, suitable for a Wikipedia article.

For example, for improving the Introduction section, the prompt:

Evaluate the language use, grammaticality, coherence, and flow of information in the following paragraph: 'Speech synthesis refers to the artificial reproduction of human-like speech. In recent years, this scientific field has witnessed significant advancements driven by extensive research, expanding computational capabilities, the accessibility of speech data, and substantial financial investments in this domain. However, the very early attempts at speech synthesis can be traced back to as early as the 18th century, reflecting the undeniable significance of speech as a primary mode of human communication. Over time, technological advancements have greatly enhanced the accuracy, quality, and naturalness of synthesized speech; see for example projects such as OLGA, which integrated audiovisual information and facial animation to boost speech intelligibility.'

returned the following feedback:

The paragraph provides a clear and coherent overview of the topic of speech synthesis. However, there are a few areas where language use and flow could be improved:

  1. Repetition: The phrase "speech synthesis" is used frequently. To enhance readability, consider using synonyms or rephrasing to avoid repetitive language.
  2. Flow of Information: The paragraph could benefit from a smoother flow of information. It starts with a general definition of speech synthesis and then jumps to recent advancements before briefly mentioning early attempts. It may be more logical to first introduce the historical context before discussing contemporary developments.
  3. Citations: When mentioning specific projects like OLGA, it's advisable to provide more information or citations to give the reader a better understanding of these references.

Here's a revised version with these considerations in mind:

"Speech synthesis, the artificial recreation of human-like speech, has undergone significant evolution in recent years. This progress is fueled by extensive research, enhanced computational capabilities, the availability of speech data, and substantial financial investments in the field. Notably, the early experiments in speech synthesis date back to the 18th century, highlighting the enduring importance of speech as a primary means of human communication. Over time, technological advancements have markedly improved the accuracy, quality, and naturalness of synthesized speech. For instance, projects like OLGA have leveraged audiovisual information and facial animation to enhance speech intelligibility."

This version maintains the essential information while reducing repetition and improving the flow of ideas.

Importantly, although ChatGPT almost always provided a revised block of text with the incorporated feedback, the authoer opted for improving the initial paragraph to the best of her abilities, considering ChatGPT's feedback selectively, i.e. the revised block of text was never copied and pasted straight from the answer prompt. Ultimately, the model's suggestions were either incorporated manually or ignored. Here is an example of partly dismissed suggestions:

The paragraph offers a comprehensive overview of the historical context and the early attempts at mechanical speech synthesis in the 1700s. The language use is generally appropriate, but there are areas where coherency and flow could be improved:

  1. Transition and Organization: The paragraph moves between different ideas and time periods with clear transitions. A more organized and structured progression of information would enhance coherency and readability.
  2. Clarity: Some sentences are quite long and complex, which can make the information more challenging to follow. Simplifying and breaking down certain sentences could improve clarity.
  3. Citations: While various historical figures and works are mentioned, citations or references to specific sources would enhance the paragraph's credibility.

Here's a revised version that addresses these issues:

"In the backdrop of the Enlightenment, the 1700s witnessed the nascent attempts at mechanical speech synthesis, reflecting the era's pivotal ideas. This historical period was characterized by a focus on human well-being, the pursuit of knowledge through rational inquiry and sensory evidence, and a fervor for freedom and scientific progress. The Enlightenment also marked a growing appreciation of the human intellect, with a strong emphasis on the scientific method and principles of reductionism.

While anatomical knowledge of the head and neck was well-established by the 17th century, there remained a significant gap in understanding how these anatomical components contributed to speech production and the precise origin of vocal sound.

A notable early account of speech production and proposed synthesis mechanisms can be found in Marin Mersenne's work from 1636. Here, he provides clear instructions for constructing organ pipes capable of articulating vowels, consonants, syllables, and complete utterances. His work, along with later contributions from Denis Dodart and Antoine Ferrein, correctly recognized the pivotal role of tube shape and its manipulation in reproducing speech sounds.

By the 1750s, the prevailing theoretical framework for speech production posited that the lungs generated an aerodynamic airflow, which passed through the vocal folds, causing them to vibrate and produce sound. This sound was subsequently shaped into speech through a tube-like vocal tract. At this juncture, the analogy between the vocal tract and a 'special musical instrument' was firmly established, and scientific attention shifted towards how to effectively use this unique vocal instrument to generate speech. Concurrently, Leonhard Euler made notable contributions to the field of physics, speculating on the possibility of creating musical instruments that could closely mimic human speech.

In summary, these contributions from various disciplines during the Enlightenment era represent a pivotal point in the early understanding of speech production and potential approaches to speech synthesis."

This revised version offers a clearer organization of ideas and enhanced readability.

Suggestion 1 was dismissed because it contains contradictory information and because I considered the timeline to reflect my intentions well and clearly enough. I have minimally revised some longer sentences; and I have completely dismissed the revised paragraph, as I think the style is excessively pompous and more convoluted than before.

Contributing student

Maria Tepei

References

  1. Suendermann, D., Höge, H., & Black, A. (2010). Challenges in Speech Synthesis. In Speech Technology. Theory and Applications. Springer.
  2. Flanagan, J. L. (1976). Computers that talk and listen: Man-machine communication by voice. Proceedings of the IEEE, 64(4), 405-415.
  3. Beskow, J., & McGlashan, S. (1997, August). Olga-A conversational agent with gestures. In Proceedings of the IJCAI'97 workshop on Animated Interface Agents-Making them Intelligent.
  4. http://research.spa.aalto.fi/publications/theses/lemmetty_mst/chap2.html
  5. https://en.wikipedia.org/wiki/Age_of_Enlightenment
  6. Casserius, G. (1600). De vocis auditusque organis historia anatomica. Baldinus, Ferrara, Italy.
  7. 7.0 7.1 Mersenne, M. (1636). Harmonie Universelle. S. Cramoisy, Paris.
  8. 8.0 8.1 8.2 Ramsay, G. (2019). Mechanical Speech Synthesis in  Early Talking Automata. Acoustics Today, 15(2), 11–19. https://doi.org/10.1121/AT.2019.15.2.11
  9. Dodart, D. (1700). Mémoire sur la cause de la voix de l'homme, et de ses differens tons. Suite des mémoires de mathématique et de physique tirés des registres de l'académie royale des sciences de l’année MDCC, 343-410.
  10. Ferrein, A. (1741). De la formation de la voix de l'homme. Suite des mémoires de mathématique et de physique tirés des registres de l'académie royale des sciences de l'année MDCCXLI, 545-579.
  11. 11.0 11.1 11.2 Ohala, J. (2011). Christian Gottlieb Kratzenstein: Pioneer in Speech Synthesis. ICPhS2011 XVII.
  12. 12.0 12.1 Story, B. (2019). Chapter 1. History of Speech Synthesis. In Routledge Handbook of Phonetics (pp. 9–32). Routledge.