Art and Community Center for Granby 4 Streets Neighbourhood
This project, set in Liverpool’s Granby 4 Streets neighborhood, re-imagines the traditional juxtaposition of old
A Glimpse Of London’s Collective Memory
London, UK / Fall 2023
We live in a world where memories are shaped by our five senses. These inputs combine to influence our personal recollections of events. Although memory is acknowledged for its complexity and diversity, the concept of collective memory remains an area ripe for exploration.
This project aims to explore collective memory by examining the interplay between visual and auditory experiences. Sometimes, the subjectivity of sound is more likely to affect people’s memories than its accuracy. As Lesley Morris puts it in “The sound of Memory”, the sound is “more elusive and perhaps more fragile and transient than the visual site of memory”. To seek a comprehensive view of how collective memories are formed, this project develops a framework that considers multiple modes of sensory input. Employing sound capture, subjective evaluation, and data analysis through advanced machine learning techniques.
The project focuses on sounds from London, which have been responsibly collected in accordance with Robots Exclusion Protocol. The data undergoes a cleansing process, allowing for the semantic segmentation and integration of text, visuals, and audio into a unified structure. This structure facilitates the creation of a controlled multi-modal input system, providing insights into the collective impression of the London city.
This project acquires its primary dataset from Sound Around You (https://www.soundaroundyou.com/). Utilizing Python, I have asynchronously harvested meta data from the London area, ensuring adherence to the Robots Exclusion Protocol. The dataset encompasses a variety of data types including audio recordings, accompanying narratives provided by the recorders, their subjective evaluations, and geographical coordinates of recording locations.
The narratives from recorders offer insights into the auditory experience, the surrounding environment, and their recording intent. To refine this textual data, I apply advanced natural language processing (NLP) techniques.
Initially, the en_core_web_sm model from SpaCy is deployed for frequency analysis of terms, rudimentary categorization, and amalgamation of significant keywords, integrating these with user ratings. I further engage the DistilBERT model, a streamlined variant of the Transformer-based BERT architecture, for preliminary sentiment analysis. This is complemented by BERT’s semantic clustering capabilities. Subsequently, content within each cluster is synthesized and abstracted utilizing the facebook/bart-large-cnn model, culminating in the formation of five distinct data clusters.
Geographic coordinates and user ratings undergo a similar clustering process. It facilitates the subsequent categorization and supervision of audio files in machine learning algorithms, enhancing the precision and relevance of our data analysis in exploring collective memory.
The analysis of the recorders’ evaluations across six criteria indicates a clear link between their perceptions of sounds and their current activities or motivations. Recorders in states of relaxation or recreation were found to be more attentive to nearby sounds, often reporting higher satisfaction levels. This tendency underscores a characteristic of London’s auditory culture: relaxing and enjoyable soundscape may influence more the collective experience and memory in urban context.
Utilizing Sentiment Analysis and Semantic Clustering, I’ve unfolded London’s map along the Thames River. This reveals a varied sentiment distribution across five categories along the river. Central areas near the Thames show increased pleasantness, with a decrease in excitement. Conversely, the Thames’ western section is notably more lively.
Sentiment Analysis & Semantic Clustering
Image Semantic Segmentation
This phase focused on visual data acquisition. Utilizing the Google Street View API, I gathered views of locations corresponding to the audio recordings. This step was essential for providing a precise visual context to the auditory data.
Following the collection, these images were processed through semantic segmentation using the ‘cityscapes-1024×1024’ model, a machine learning model adept at urban scene analysis. This advanced segmentation technique categorized key visual elements within each scene, such as buildings, roads, and natural landscapes, offering an in-depth visual analysis of each recording location.
Complex Syntax Integration With Auditory Labels
Audio analysis involved extracting spectrograms from select recordings for manual labeling, enhancing the algorithm’s pattern recognition for different elements in each cluster. The combination of descriptions, text sentiment data, geographic locations, ratings, image semantic segmentation data, and auditory labels collectively contributes to a complex syntax.
Influence of Complex Syntax
The complex syntax employed in the framework significantly influences the expression of images generated by artificial intelligence. AI models such as Dall-E, Mid-journey, and Stable Diffusion, trained on the guides of the syntax, are adept at capturing the nuanced impressions of London. While these images are occasionally surreal, effectively mirror the subjective emotional dimensions of collective consciousness. This process underscores how intricate data manipulation can shape AI’s visual representations, offering a deeper insight into the collective memory and sentiment towards the city’s diverse auditory and visuals.
Collective memory, inherently rich in its chaotic complexity and multiplicity, presents a unique mix of communal experiences and emotions. Use monocular depth mapping to transform this disordered aggregate into a structured, visual form. This visualization acts as a medium of imagination, transcending the conventional boundaries of memory representation. It may offer a methodological approach that opens up new possibilities in design and interpretation. In this realm, collective memory is not just a recollection of the past but a dynamic interface that melds historical perspectives with contemporary evaluations and future outlooks.
Interface Of Memory Manifestation
This project, set in Liverpool’s Granby 4 Streets neighborhood, re-imagines the traditional juxtaposition of old
This project explores the application of pneumatic formwork construction technology combined with robotic arm assistance