4
Evasion Scores


Evasion Scores is an iterative project that investigates relationships between the voice, technological surveillance, and the politicisation and militarisation of words and phrases in everyday speech. This project is an ongoing response to the rapidly increasing prevalence of domestic surveillance devices that capture and commodify vocal data. 
                            The project is presented as graphic scores, performance lectures and recorded performative demonstrations to propose ways of using the voice to evade detection from domestic smart speakers and other machines that listen. Various extended vocal techniques are used to prevent smart speakers from 'listening' to and understanding speech. The idea is that these vocal techniques can be incorporated into everyday speech; adapted, built upon and used iteratively in human-human interaction.

Evasion Score I
Utan Evasion Score 1: Utan introduces this ongoing project with a simple score of ten vocal lines. Each line is a sentence or phrase that I have used or heard used by friends or family in my private life. They are seemingly innocuous phrases that in the wrong context, could signal transgression against state authority, political or philosophical/moral subversion.
                            I applied a series of vocal techniques to each line and tested them out on my iPhone 12 speech-to-text function, and an Amazon Echo dot, to see what each machine could hear. 
                            The subheading of this Evasion Score is called 'Utan', an American fake tanning product, because my phone continuously suggested that this is what I was trying to say, when it couldn't understand me.





Evasion Score II
Domestic Security As a starting point for this graphic score and video performance, I used a corpus of words released by the USA's Department of Homeland Security, under a Freedom of Information request issued in 2012. 
                           The 331 words listed, are actively monitored on social media platforms as triggers for potential terrorist or other threats against the USA. The list includes words such as "bomb", "terrorist" and "attack" as well as more pedestrian words such as "exercise", "relief" and "worm".
                           To construct the score, I took a selection of keywords and placed them in quotidian/ everyday speech sentences.
                            I then developed a register of vocal techniques that can be applied to the keywords used within the score, in an effort to render them unintelligible to machines that listen, but still possible for human ears to discern.
                           Finally, I made a recording of the score and uploaded it to REDUCT, a video transcription service that uses a neural network to listen to videos and transcribe spoken words into text. 
                            The resulting video is captioned with REDUCT's transcript, demonstrating instances of slippage, where the vocal techniques have managed to render the keywords unintelligible altogether, or where they have been misinterpreted as other words.  




Evasion Score III
CallHOMEIn 1997, the Linguistic Data Consortium called for native English speaking members of the U.S. public to participate in an early project on Large Vocabulary Conversational Speech Recognition (LVCSR), sponsored by the Department of Defense. Speakers were sourced via the internet, publications (advertisements), and personal contacts. The LDC collected and transcribed 120 unscripted 30-minute telephone conversations. Most participants called family members or close friends. The participants were made aware that their telephone call would be recorded, as were the call recipients. The call was allowed only if both parties agreed to being recorded. Upon successful completion of the call, the caller was paid $20 (in addition to the call being toll-free). The resulting corpus of conversational speech was an early milestone in developing the natural language processing, machine learning and artificial intelligence technologies we are familiar with today.
                            For 'Evasion Score 3: CallHome', I acquired and downloaded the recordings and transcriptions from the CallHome English speaking corpus. The conversations are banal and often intimate, and I selected excerpts that reveal important and poignant information about the participants' lives, emotional states, hopes, desires and daily activities.
                            I devleoped a register of vocal techniques designed to render specific words and phrases unintelligible to machines that can now listen (thanks to the very utterances of these words 20 years ago), such as domestic smart speakers, smart phones and other speech recognition devices.
                            The score proposes a way to re-perfrom these everyday conversations within the context of contemporary surveillance capitalism.

---

←4

→