Jacque Fresco was a pioneering designer and futurist whose visionary concepts challenged conventional notions and offered a glimpse into a future characterized by sustainability, resource-based economies, and technological innovation. Jacque Fresco was not merely a thinker; he was a visionary whose insights transcended conventional boundaries and offered a compass for a more compassionate, equitable, and harmonious world. Through his thought-provoking lectures, Fresco challenged the status quo, urging us to envision and strive for a future guided by science, innovation, and readiness for change.
Hence it is with tremendous gratitude that we announce the completion of the transcriptions project of Jacque Fresco’s vast collection of lectures. This achievement stands as a testament to the enduring legacy of a true pioneer, whose ideas continue to shape and inspire generations across the globe. Jacque Fresco’s visionary ideas in sustainable design and economics have captivated and influenced minds for decades. His influence has inspired many to undertake similar work, drawing from just a fraction of his extensive lecture collection. The excitement lies in the anticipation of what might emerge if individuals engage with over a thousand of his lectures, recognizing the transformative potential of these insights for all those willing to delve into Jacque Fresco’s thinking. This remarkable achievement has been realized by the dedicated transcription team led by Nathanael Dinwiddie.
The Transcription Project is one component of a broader program to archive Jacque Fresco’s body of work for posterity. This work began in 2011 when Nathanael Dinwiddie commenced an exploration of Jacque Fresco’s biography and creative works. This exploration included surveying the assets in Fresco’s possession, canvassing all sources of additional information and source records pertaining to Fresco’s history, and contacting all associates who suspected of being in possession of materials derived from Jacque Fresco. Early work included review and organization of physical assets, subsequent collation and systematization of metadata in databases, and finally digitization of all physical items which includes:
- 616 audio tapes
- 1,879 video tapes
- 195 publicity documents
- 5,500 design sketches
- 200 schematic designs
- 400 handmade scale models
- 297 books
- 85 hard drives
- 113 floppy discs
- Hundreds of CDs
- thousands of photographs and film negatives
- thousands of text documents
Informally initiated in 2015 and inspired by Transcribe Bentham, the Transcription Project formalized in 2017 when a systematic workflow and guideline was created to guide volunteers to completion and a team was created on The Venus Project’s Trello work platform. There were many attempts to automate the process through machine transcription but the results were always disappointing. It seemed the work should remain a human task until transcription algorithms improved. Although it began with a slow start and working through volunteers, it received a boost in December 2019 through a fundraiser that allocated 25k to purposes of transcriptions.
A prior survey had identified 19 transcription companies, the cheapest of which would cost 202k to transcribe all lecture content of Jacque Fresco and would extend across multiple years. It was also found that hiring freelancers could reduce the cost of transcription to as low as 80k. This high cost prompted a prioritization of lectures suspected to offer the most value (unique content). The number of lectures was reduced from nearly 3,000 (at over 3,000 hours) to 864 (around 1,800 hours) at a cost of 62k for the cheapest commercial company or 25k for freelancers. It was evident that freelancers were the cheaper option. Thereafter, freelancers were hired from the freelance website, Fiverr.
Freelancers from developing nations with foreign languages were the tradeoff to the low cost. Because it was essential to attain the highest quality of results in each transcription to reduce the later burden of correcting errors in proofreading, an extensive survey was conducted to identify those individuals most capable of accurately transcribing Fresco’s lectures despite transcribers speaking a foreign language. Those found to transcribe most accurately were from Jamaica, India, Kenya, and Nigeria where a dialect of English is commonly spoken. The survey considered all freelancers classified at “Top Rated” or “Level 2” in Fiverr’s ranking system. The survey filtered 640 candidates, reducing the number to 79 who met criteria. These candidates were sent a low quality audio sample of a Fresco lecture in order to find who could best transcribe the difficult audio. Of these 79 candidates, 31 transcribers had an accuracy above 90%. Of these 31, 20 were hired to transcribe at an average cost of $0.50 per minute of audio. Now the challenges started. The following issues arose:
- Formal withdraw because the transcription was too difficult
- Unreliability of delivery, delays or incompletion
- Subcontracting other transcribers of lower quality
- Unannounced usage of AI transcription tools
- Loss of contact during COVID pandemic
Some transcribers hired other people to do their work for them (subcontracting). This changed the quality of the text. They were asked to refrain from subcontracting. The usage of AI transcription tools was prevalent. To catch these deviations in quality, it required checking the quality of text to assure the transcribers methods had not changed. On multiple occasions transcribers were asked to not use such tools. At the time of 2020-2022, algorithms for audio transcription were not reliable for low quality audio and introduced many errors that proofreaders would need to later correct. Machine transcription was especially poor performing with regard to punctuation. Often, the quality was so poor, it justified creating a new transcript from scratch via human transcription.
Upon completion of a transcript, it was reviewed and bulk corrected to eliminate any common formatting or orthographic errors, then submitted to a pool for volunteers to proofread. An app was constructed to better facilitate volunteers accessing lectures and updating the progress of their work. During COVID, transcriptions had slowed to halt.
In September 2022, Open AI released Whisper speech recognition system. Performance of this tool was assessed against the samples given to transcribers and found that Whisper scored higher than the best transcriber at 96% accuracy. Thereafter, experimentation began with Whisper in combination with algorithms for audio enhancement to test its capability to transcribe the range of audio quality in the Jacque Fresco audio collection, including the worst quality audio for which it still performed well. It was found that Whisper performed best when no adjustment was made to the audio via sound editing. Ultimately, Whisper 3 was used to transcribe all remaining lectures on the priority list in addition to the entire inventory of Jacque Fresco’s audio content which comprises over 2,959 unique recordings (lectures, interviews, discussions) across 4,240 audio files at a total of 3,200 hours.
How did we do it? The lectures were transcribed using OpenAI’s Whisper large-v3 model. They were processed on Runpod’s cloud GPUs on Nvidia 4090 GPUs. Labeling of speakers was achieved by first segmenting the transcripts by ‘meaning’ using AI21’s text segmentation models. After segmentation, we discovered that it would cleanly separate speakers in a conversation by line breaks. We then aligned Whisper’s word-level timestamps to each segment, yielding segment-level timestamps. Those timestamps were used to split the audio file up to get the corresponding audio segments where a single speaker was speaking. Then each audio segment was embedded using the speaker embedding model (wespeaker). Embeddings are compressed neural numerical representations for data. They can be likened to faces. Some faces are similar, and embeddings give us the ability to measure how similar two data points are. Audio segments’ 256-dimensional embeddings were then dimensionally reduced using UMAP to 2 dimensions and clustered using HDBSCAN. A cluster is a group of data points which are similar. Each cluster was reviewed by ear to identify the corresponding speaker and label them accordingly. Finally, speaker labels were applied to each segment programmatically. Figures 1-3 provide examples of how the speakers cluster into distinct distributions.
With AI’s help, the transcription team accomplished this incredible feat at 54% of the initial budget. The remaining funds will fuel other vital projects within the Jacque Fresco Foundation, including paving the way for the formalized publication of Fresco’s rich and complex body of work. Imagine such ideas reaching even further, igniting interest across the globe! For this purpose, the transcripts still require extensive proofreading.
The completion of this transcriptions project is a triumph of dedication, collaboration, and unwavering commitment to preserving Fresco’s invaluable contributions to humanity. Volunteers, scholars, and enthusiasts from diverse backgrounds united in a shared mission to transcribe and proofread his lectures with meticulous precision, ensuring that every word, every idea, and every nuance is faithfully captured and preserved for posterity.
Fresco’s lectures serve as a treasure trove of insight, offering readers a profound journey into the mind of a visionary thinker whose ideas remain as relevant and urgent today as they were when he first shared them. In these transcriptions, readers will encounter a wealth of transformative concepts—from resource-based economies to sustainable urban design, from holistic education to the role of technology in shaping our collective future through navigating the complexities of everyday life. Fresco’s lectures are not merely intellectual exercises; they are beacons of hope, guiding us toward a future where inquiry, creativity, and compassion reign supreme.
As we celebrate the completion of this transcriptions project, we also reflect on the profound impact of Jacque Fresco’s legacy. His ideas continue to resonate with individuals and communities around the world, igniting imaginations, sparking conversations, and inspiring proposals aimed at realizing his vision of a better world for all. Hence, we look forward to making Fresco’s lectures and these transcripts available in the future as part of the mission of the 501(c)3 nonprofit organization, Jacque Fresco Foundation.
To all who contributed to this monumental effort, we extend our deepest gratitude. Your dedication, passion, and unwavering confidence in Jacque Fresco’s vision have ensured that his legacy will endure for generations to come. Let us honor his memory by embracing his ideas, fostering dialogue, and taking meaningful action to build a future that honors the best of humanity’s potential. In the spirit of Jacque Fresco’s vision, let us continue to envision, innovate, and collaborate toward a world where wisdom, sustainability, and collective well-being are the cornerstones of our shared reality.