Jack Henrie Fisher » lectures

Transcription of the voice as the image of thought

admin — Thu, 18 Feb 2010 15:27:59 +0000

1
I am talking now in a counterpoint to another speech, arranged as lines.
I am talking through a metrical order: this is a talk about how to write, while listening to another’s voice.
On the way to talking about some recent experiments (I am talking as a graphic designer) — experiments with transcription, with the encounter the formal can arrange within the discursive — I find too a chance to develop an aleatory formula for constructing a thought, in the gaps of, from the material of, speech: from the gaps in the material of another’s speech.
My research began, last year my intention was to make procedures for indexing public speech at Jan van Eyck: to index everything that said, to assemble a machine, a machine to read through, to index and disassemble, a stable archive of speech — a body of speech automatically transcribed by a computer. But a technical failure occurred at the very beginning: the computer’s conversion of speech to text produced on most occasions 
2
either nonsense or silence.
The computer’s transcription failed precisely because of the polyphonic conditions, the non-standard intensities of the vocalized international English of the Academie.
The machine stumbled on the voice.
In the advent of this failure, my attention was forced to the site of the input: to the material of the voice as it can be made into a text. As the copyist of a rapidly growing audio archive, of talk after talk, I had to discover a way to quickly produce a text from another’s speech.
Under duress, under a coercion, acting under the constraint of my design problem, in circumstances where urgency demanded destroying the stages of problem-solving progression, I chanced upon a technique for doing this work. I’m calling this technique metrical transcription. Metrical transcription is a technique for producing a text while someone is speaking. The technique is founded in a cut: in the cut made in copying the words of a speaking voice.  It is the binding of the voice to a mnemonic segment, to what set of words can be remembered together. It’s therefore a variable measure, but still: a line, a verse, a chain.
The interval is non regular. It is what could be called a non rhythmic meter; or it is a rhythm not measured by quality or quantity of sound, but by something else: by the attention under constraint.
The axiom from which the technique of metrical transcription proceeds is the time of the transcription is equal to the time of the transcribed. The transcription can only take place 
3
within the time of the speech. The speech cannot be paused. It is copied exactly in the time it is spoken. The line is cut at the end of what can be remembered together. It is in this cut of the attention, as line follows line, that something else appears between them, another intention: an effect of the concatenation.
The metrical procedure is first and always as transcription a literal adherence, a fidelity to what is spoken, to what is spoken  especially without the support of writing, i.e. to improvised speech. It is a submission to the dictation of the spontaneous flight of thought. Every word that is copied is literally the word of the speaker.
Under this constraint I (and others) have produced a set of 15,000 lines for the year 2009. The unabridged edition of this incomplete concordance will be offered to libraries, to the future philologists and exegetes, the analysts of what was spoken here.
In the metrical cut of these lines, every asemantic element disappears. There is no punctuation in speaking, only words and sounds, but sounds for now are another story. When a machine hears a voice speaking —the machine is the irresistible figure for this activity 
4
when the machine copies as words the sounds of the voice, it cannot add a comma or quotation mark. It cannot add a question mark, and it cannot add the word cough or the word laughter. There are no stage directions, no punctuation, no orthographic variation in the transcriptions. Words too are left out in the exacting metrical cut. More speech disappears than is written. The writing is always a technique of reduction, an elimination: a push to sobriety.
A question is, the technique opens the question: what fastens the words together, what minimally constructs them as a chain, why do some words rather than others appear and stick together in this writing-while-listening?
The computer failed to produce sense from speaking, but the full, the complete conversion of speech to text, even in the most human, in the most non-automated transcription practice, is never self-evident. Any transcription is founded in an impossibility at the most molecular scale.
Take for instance the unprinted space between words. In the regular blank spaces of writing and printing, even the most accurate transcription with a conventional alphabet can’t register speech’s uneven movements between words. The pauses and slurs, the everyday rubato, as John Cage has called it, is rendered in the relatively fixed unit of the typographic space, 
5
just as sounds are rendered in the fixed units of letters, which constitute a separate order, which Emil Ruder, eloquent and terse ideologue of modernist typography, defines as that which can be cast as often as necessary, as that which always goes on being repeated in a precise and invariable form. There is a cool and fascinating beauty about printed letterforms, he says, they are free from the alien borrowings of speech.
The radical incommensurability of speech sound and letter form, which is declared in this procedure, opens an entirely other logic of signification. These experiments will seek to return their precisions to the exigencies of speaking. How?
The copyist must always make a mark, an interpretation from the point of this failure, from the incommensurability of these orders. He must coldly formalize speech as type. And to make speech into prose, into periods of sense, the copyist must go further, and add and switch the letterforms — must punctuate and even italicize speech into sense.
An experiment to intervene before this normalization, metrical transcription is an ascetic technique: it is a training of the attention through constraint, through the most minimal procedure of writing. The moment of transcription is the dangerous moment, Derrida writes in another context, 
6
the copyist is always tempted to add supplementary signs to improve the restitution of the original: the good copyist must resist the temptation of the supplementary sign.
It is a surprisingly difficult task to hear and copy the word exactly, when listening to improvised speech, without a supplement, to not correct, to not fill in what wasn’t said, but expected, to not punctuate, to not ignore the stuttering and anticipation, the slips of spontaneous speech. A strange effect of this training is the obviation, even the prohibition, of the necessity to understand what is being spoken. It’s actually a relief: there is a relief of the necessity to understand. We’d have a much better track, Gene Hackman’s character admonishes his assistant, in The Conversation, if you’d paid more attention to the recording and less to what they were talking about. The big picture of what is being spoken must recede, in order for the writing to take place. The drift of the meaning, the drive of the argument is lost in the granularity of the speech, in the sound of the voice and the molecular sense of the words.
I want to say too, it’s not just spoken language, in the argument of this procedure — its errancies, its stuttering, its vernacular — that is being privileged at the expense of writing. It is rather the curious and always deadpan
7
the cold interpolation of speaking into an alphabet, the acousmatic effect of the speaking voice in its registration as letterforms, which opens a new resonance for discourse as such. A so-called good or faithful transcription always presupposes specific knowledge. What is being talked about? What is the name that was said? What language is being spoken? Is that a citation? Speaking blurs the word. There is never in the first place a stable or differentiated corpus of speaking.
Homophonies for instance abound. Is that the word here or the word hear? Affect or effect? Err or air or heir?
And there is a lexical territorialization proper to any discourse. The transcriber who already knows it can more faithfully record the meaning of the speech. But what happens when we surprise the discursive field of the voice as an outsider, as an interested outsider, in words used to describe Ray Johnson — as an outsider with a will to respeak but with nothing perhaps to say. What is captured from this position when the quarry is not the meaning but the vocabulary, the word, the phrase, the atomized materials of meaningful speech?
In the writing of the naive transcriber, what sense shows through the metrical protocol? What semantic atmosphere is produced in the procedure? How can this partial record, all the successive vibrations, the bits of speech which the listener, the subject,
8
the subject of a listening articulates within the expansions of the voice’s sonority —how can these disconnected lines, the transcription, be sutured again? Into what kind of sense? What procedure, what punctuation, what is a logic to produce a rewriting, what will could connect them?
For today I want to re-present the lines of the transcription. I want to reproduce them, to condense them as a new catena, as chains of commentary. I want to digest the transcription and reconstitute it. This is a first essay: how can it be rewritten? It is submitted to a precise aleatory and semantic recombination. I want to put the lines into stanzas, into six line stanzas, as if they might begin to behave as a sort of sestina, that most programmatic of poetic structures, but a sestina that links not to itself but to another text, a text outside it: this one, the one i’m speaking.

Partial Phoneme

admin — Thu, 02 Jul 2009 08:25:43 +0000

hm(m)

The Machine Stumbled on the Voice

admin — Thu, 02 Jul 2009 08:20:57 +0000

Speech, vocalized, is disarticulated by the air. Words resound as sonic intensities just as they become empty indexes of an event.
From the machine to the voice: to the voice as exactly that which makes a rupture in the midst of an indexing machinery which is built on the legibility of the text.
The voice as a surplus to, but also as the cause of the text—as that element which always eludes the capture of sense but which at the same time founds an authority of meaning.
—The sound of the voice as the material of asignifying intensities, a sound that cannot be fully inscribed as a text but can instead be traced as linking sets of disconnecting paths and circling divagations.

(And: transcription of the voice as the image of thought which is a contingent becoming. There is a sense in which the true and elusive machination of thought inscribes its presence in the asymmetrical circles of our talking—in the foundering and hesitation—in the jumping forward …backtracking, which nondiscursively marks our talking to another. —And in the often unconscious rhetorical mechanism (prosopopeia) by which one voice inhabits another, as citation and indirect speech. I am pursuing this intersection of the vocal and the machinic as the image of thought. I am trying to find, to reveal in its occlusions, the hidden and machinic lever of thought which is embedded in the rupture of the voice. “The search for the voice in language, this is what is called thought.” (Agamben quoted by Nancy 2002, p.45))

While there is a clear psychoanalytic context for thinking through the sense of the voice-as-object (as objet petit a), my sense of the voice-as-rupture equally describes the semantic economy of the internet: a voice cannot be searched—it does not yield to the indexing machinery of the search engine. It poses therefore a void in the heart of contemporary networked productions of meaning. This is the rupturing event—the voice as void—I am tracking in all its negative singularities.

Wikipedia entries, by date

admin — Fri, 12 Jun 2009 11:56:58 +0000

1.15 Enigma Machine
1.16 Cybernetics
1.16 Artificial Intelligence
2.2 XML
2.2 XSL-FO 1.0
2.10 Unicode
2.10 Camel Case
2.10 Demonstratives
2.10 Deictic
2.12 Las Meningnas
3.19 Prosody
3.21 Autopoesis
3.23 Conversation, Bohm Dialogue
3.24 Enthymeme
3.24 Basic English
3.26 Asterism
3.27 Interview
3.27 Nevermind It’s An Interview
3.28 Transclusion
3.30 Compiler
4.1 Computational Linguistics
4.2 Audio Search Engine
4.2 Disambiguation
4.10 Relational Database
4.14 Google Search
4.15 Unicode
4.15 Clinamen
4.15 PunCtuation
4.28 Free Association
4.28 Speech Corpus
4.28 Man Pages
5.17 Vocal Folds
5.30 Free Indirect Speech
5.30 Prosopopeia
6.02 Oracle
6.02 ESP
6.05 Grammar
6.05 Encylopedia
6.10 Open Document

Hearing Writing

admin — Mon, 16 Mar 2009 13:02:25 +0000

Imagine the following model for writing.
Every text exists first as something read aloud by another. If a text is authored, if it is original, it must be vocalized by its author; it cannot be written down. A text, as writing, can only exist as something copied by a listener in the time in which it was vocalized by its author.
Any text can be produced only as a partial transcription of the speech of another.

Transcription Constraint

admin — Wed, 18 Feb 2009 18:37:00 +0000

It is very evident early on that any transcription will be produced by virtue of serious exclusions and distortions. An early question I am having is to what degree to intervene in the transcription, in order to produce something coherent and readable.
If the transcription cannot be produced accurately by software from audio recordings, then discriminating strategies of transcription must be developed in a human agent.
In this scenario it is readily admitted that the transcription will be partial. The rule is as follows: the time of the transcription is equal to the time of the transcribed. That is, the transcription can only take place within the time of the speech.
Within this constraint, how does the scribal agent operate? What further rules constrain his transcription? In this picture, is he a machine or does he function more as an antenna?

Index Machine

admin — Wed, 18 Feb 2009 18:22:51 +0000

The project is to capture speech and to transform it into writing according to a developing system of constraints.
The project is—at the same time—to develop a space and set of tools by which a community can author a collective statement.

And: there is an emerging interest in this research to document the very process of research itself. That is, there is an interest germinating in the scribal agent (jack) to map within the text-repository every citational movement and articulation of knowledge connected to the subject, which is: the machinic enunciation.
This is the surplus activity of research, conducted by the scribal agent, which occurs alongside the action of transcribing.

Script—Speech—Subtitle

admin — Sun, 15 Feb 2009 10:10:18 +0000

In a film for which subtitles in another language are provided, the subtitles are presumably derived from the script which the actors follow, if there is one. This is evident in certain moments of a film, for instance when the speech is too quiet to be heard but still receives a subtitle. Thus the subtitles are not a register of what is spoken, but belong to the inscription which exists prior to its being-spoken.
How might this model be made to impact the conceptualization of my transcription project? That is, what would it mean if the often garbled and nonsensical text which is called the transcription is regarded, in a reversal, as the script which the actors at the Jan van Eyck are speaking—and diverging from in order to bring their speaking into “sense?”

Formal 1

admin — Mon, 02 Feb 2009 20:35:24 +0000

What is the minimum of typographic form necessary to legibly reproduce a text and/or to write? Italic, the underline, a space between paragraphs.
There is a possible book in which italics, say, do several kinds of work: to indicate citation, proper name, but also (at the same time) as an emphasis.
What would it mean to collapse these semantic orders into a single distinction?

Argument for Typography 1

admin — Thu, 29 Jan 2009 15:08:18 +0000

Turing reformulates the question, Can machines think?, in the terms of a game in which an interrogator questions two (invisible) entities in order to determine the gender of each [Turing, 1950]. The question becomes whether the interrogator will decide wrongly as often when one of the entities being questioned is a machine.
It is critical in this game that the answers be given as printed, as typewritten, rather than vocalized or written by hand. “The ideal arrangement is to have a teleprinter communicating between the two rooms.” The printed letters eliminate traces of gender. Typography furnishes the condition for machinic masquerade.