Selected Projects

Recursive Dictation

2013-4, multimedia installation w/subsidiary texts

Beginning with version 10.8 (July 2012), Apple Inc.'s Mac OS X desktop operating system has included Dictation, a feature modeled on Siri, the iPhone voice-recognition interface. When activated, this feature cross-references spoken audio input against an online database of speech data (including a log of the user’s previous speech input) and, using this information, transcribes the audio into text. The accuracy of this transcription is variable. Mac OS X also features a speech synthesis engine (“Text-to-Speech”) capable of “reading” text selections aloud in a variety of differently accented computer voices. The verisimilitude of these various voices is, likewise, variable.

In Recursive Dictation, a specially written computer application recursively routes Mac OS X Text-to-Speech output to the Dictation feature and Dictation output to the Text-to-Speech feature... ad infinitum. The result of this process is an iterative stream of text and synthesized speech that, due to the limitations of speech synthesis and speech recognition software, is both ever mutating and never-ending.

Because Siri is designed for utility and everyday use, its results are weighted towards words that meet the application's definition of vernacular speech. This definition is fluid; it adjusts in real-time as new instances of user interaction are uploaded to a centralized database and analyzed. Recursive Dictation shines a light on this aspect of the Siri system, illustrating how the application's speech recognition database functions as an up-to-the-minute ephemeral snapshot of the practical vernacular--an ongoing archive gleaned from countless--otherwise overlooked--moments of quotidian human-machine interaction.

Primarily, the intent of Recursive Dictation is to explore the creative potential of misunderstanding and (mis)translation, specifically with regard to computational linguistics, natural language processing in particular. Yet, in tandem with this expressive aim, Recursive Dictation can simultaneously be understood as operating in a more explicatory register, serving as an exemplification of the challenges faced by researchers working in those fields. For instance, because the Recursive Dictation application allows for adjustments both in the rate of synthesized speech and in (simulated) regional accent, its output offers a demonstration of how these variables affect the success of computerized transcription. Further, through the repeated playback of speech synthesis samples decelerated by such adjustments, the application makes vivid the gulf that still exists between synthesized speech and the nuances of human vocalization. Though worthy of consideration in and of themselves, such technical issues take on increased significance when examined in the context of disability and the manufacture of effective assistive devices.

Excerpt: "Cumulative error..."

Excerpt: "This text is an application..."

Subsidiary Pieces

In addition to functioning as a work in itself, the stream generated by the Recursive Dictation application also serves as a foundational element in numerous subsidiary pieces — chiefly, poetic texts intended for print and live performance — that likewise fall under the Recursive Dictation umbrella title. These compositions elaborate the issues of translation central to the main work by reinterpreting particular executions of the application stream as, essentially, unannotated performance scores. Thus, in these works, syntactic and semantic deficiencies in the application’s output are approached as interpretative challenges, formal constraints negotiated by means of prosodically modulating the text so as to imbue it with fixative layers of connotation. The three recordings below are representative examples of this reading practice.

  1. Good Morning All (excerpt) Play Pause 00:00 / 01:55

  1. Get a Visual Play Pause 00:00 / 00:51

  1. The Biting it Beats Out Play Pause 00:00 / 00:25

Copyright © 2016 Christopher Vandegrift