วีดีโอ

Brain plasticity really does wonders allowing our brains to learn anything, including when watching youtube videos for free th-cam.com/video/KQ3dOn0hWWg/w-d-xo.html "need to really show what an atypical EEG looks like, and then show that training those patterns improves the conditions they’re supposedly associated with." p.s. wish the admission of being a charlatan, came in the beginning of the interview, not the end.

Hello sir, I hope you are well. I saw that your TH-cam videos are very lovely and very helpful. You have enough of the videos but your video is not getting views because there is some problem with your videos. Your SEO score is very poor (00% out of 100)and you need optimization for better results. If you update the videos then the views of the video will increase. If you want you can discuss with me. Thank you so much.

12:48 how did you remove the dog howling in the background? th-cam.com/video/YSnqJQC1eEw/w-d-xo.htmlsi=YxHXoDIl3Xh0XWPn&t=787

Why did you choose 720x720@25 frames for this video? original was 1280x720@30

❤❤❤❤

First

When participants engage in internal speech, do they naturally shift their gaze toward the word on the screen or paper, and could these subtle eye movements play a role in selecting or reinforcing the word they are thinking about, even though such movements are unlikely to be detected in S1?

1:25 - Translating Brain Signals into Speech 1:36 - Publication on Decoding Internal Speech in Human Cortex 2:00 - How Brain-Machine Interfaces Decipher Internal Speech 4:05 - The Role of Neural Interfaces in Neuroprosthetics 5:12 - Decision to Study Biomedical Engineering over Medicine 5:46 - Introduction to Neuroprosthetics and Bi-engineering Studies 6:25 - Early Work with EEG and EMG in Neuroprosthetics 7:42 - Transition to Advanced Brain-Machine Interface Research 11:01 - Decoding Internal Speech with Intracortical Arrays 13:19 - Research on Decoding Internal Speech vs. Vocalized Speech 14:10 - Approaches in Speech Decoding Using Brain Areas 16:02 - Internal Speech Decoding for Locked-In Syndrome Patients 17:22 - Control Experiments on Internal Speech Mechanisms 20:05 - Speech Representation in Brain Areas Through Different Modalities 22:00 - Brain Responses in Silent vs. Vocalized Speech 24:07 - Experimental Setup for Presenting Words to Participants 25:09 - Process for Training Decoders with Limited Data 27:21 - Participant Strategy and Neural Encoding of Internal Speech 28:46 - Signal Processing Techniques for Decoding Speech 29:55 - Electrode Array Configurations: 64 and 96 Electrode Arrays for Neural Recording 30:01 - Electrode Types and Surgical Implantation: 64 vs. 96 Electrode Arrays, and Longevity of Implants 31:22 - Daily Variation in Active Units: Range of 30 to 60 Units Recorded from Each Session 33:05 - Classification Techniques and Decoding Word Accuracy 34:07 - Real-Time Decoder Training and Feedback 35:04 - Advancements with AI and Language Models in Speech Decoding 37:03 - Comparison to Previous Studies Using EEG and ECoG 38:06 - Real-Time Decoding Achievements in Internal Speech 40:15 - Main Findings: Real-Time Internal Speech Decoding Capabilities 41:03 - Challenges in Verifying Internal Speech Decoding Validity 43:20 - Impact of Electrode Number and Quality: Variations in Decoding Accuracy Due to Fewer Electrodes 44:50 - Language and Phonetic Representation in Neural Data 46:05 - Future of Speech Neuroprosthetics and Neural Technology 50:25 - Ethics and Concerns in Internal Speech Decoding Technologies 55:07 - Differences in Speech vs. Grasping Data Processing Techniques 57:11 - Low-Frequency Potentials (LFPs) in Speech Processing 59:04 - Insights into Data Quality and Vocabulary Expansion 1:00:04 - Advancing Internal Speech Models with Increased Data 1:07:05 - Key Question: Practical Challenges in Decoding Speech 1:08:01 - Resources: Access to Code, Data, and Study Reproduction 1:09:50 - Importance of Making Neurodata Accessible for New Researchers

Hope the gear and electrodes recording better signal to noise ratio than the microphone used in the video

Thank You

In cochlear implants, increasing the number of electrodes beyond approximately 22 does not significantly improve sound perception. This is because the brain’s auditory system can interpret sound effectively with a limited number of frequency channels due to the linear organization of the auditory nerve and the brain’s adaptability. In contrast, speech BCIs require far more electrodes (hundreds to thousands) to capture the complex neural signals associated with speech motor control. However, similar to cochlear implants, increasing electrode density in BCIs faces diminishing returns. Beyond a certain point, adding more electrodes may not meaningfully improve accuracy due to signal overlap, redundancy, and the brain’s biological limits in generating distinct signals. Given these similarities in diminishing returns, how do you think we should balance the number of electrodes with the practical limitations of processing and interpreting neural data in speech BCIs?

12:33 - Introduction to Speech BCI: What is a Speech Brain-Computer Interface (BCI) and how does it work? Overview of decoding speech from the human cortex to help people with speech loss due to paralysis. 14:26 - Ideal Speech Neuroprosthesis: the three main goals of an ideal device: long-term stability (decoding over decades), high accuracy (97-98% accuracy for large vocabularies), and non-invasiveness (avoiding brain surgery). 17:48 - Milestones in Speech BCI Development: The lab’s use of electrocorticography (ECoG) to decode movements of the vocal tract, jaw, and tongue to synthesize speech directly from brain signals. 19:02 - Speech Decoding Using AI: Explanation of the 2019 study where AI models were used to decode speech from ECoG signals. Key insights into how AI techniques used in speech recognition on phones (e.g., Siri, Alexa) were adapted to predict text from brain activity. 22:02 - High-Performance Neuroprosthesis (2023 Study): In-depth discussion of the 2023 study on speech neuroprosthesis, which involved decoding speech in a participant with paralysis. Use of ECoG grids with 253 channels to decode over 1,000 words with a 25% word error rate. 22:56 - Large Vocabulary and Accuracy Challenges: How the lab expanded from small vocabularies (50 words) to over 1,000 words. Use of AI models to manage open vocabulary decoding with real-world applications. 25:29 - Voice Reconstruction with AI: Detailed breakdown of the process of reconstructing a patient’s voice using acoustic units. How a participant’s wedding video was used to synthesize her original voice by transforming brain signals into waveforms. 27:05 - Overcoming Temporal Alignment Issues: Use of Connectionist Temporal Classification (CTC) to deal with the problem of aligning brain signals with speech outputs in paralyzed patients. The importance of this algorithm in automatic speech recognition and BCI research. 28:45 - Increasing Electrode Density for Improved Decoding: Explanation of how increasing the density of ECoG grids (253 channels) improved the spatial resolution of brain signals and led to more accurate speech decoding. Future potential for 512 or more channels to improve performance further. 31:31 - Combining Text, Speech, and Avatar Control: How the research team integrated text decoding, voice synthesis, and facial movements into one cohesive system. Driving a high-fidelity digital avatar with brain signals to simulate natural facial expressions and speech movements. 34:05 - AI and Speech Decoding: How modern AI models, originally developed for speech recognition, were adapted to predict text and acoustic units from brain signals. The benefits of machine learning models in handling large vocabularies and increasing accuracy. 37:03 - ECoG vs. Microelectrode Arrays: A comparison of ECoG and penetrating electrodes (microelectrode arrays). Discussion of the trade-offs in terms of invasiveness, signal stability, and spatial resolution. The role of ECoG in long-term, stable recordings for speech decoding. 40:58 - Long-Term Stability in BCIs: The lab’s focus on ensuring long-term stability of ECoG grids over 5-10 years or more. Challenges with penetrating electrodes like Utah arrays, which suffer from immune responses and signal degradation over time. 47:27 - Improving Accuracy with Increased Channels: Future directions for increasing the number of electrode channels in ECoG grids to improve decoding accuracy. Discussion of how performance improvements are seen with every increase in channel density, leading to better resolution of neural signals. 49:36 - Future of Speech BCIs and Non-Invasive Approaches: Exploration of potential non-invasive alternatives to ECoG, such as Neuralink and Synchron’s Stentrode, which could offer less invasive ways to achieve high-resolution neural recordings. Challenges and promises in non-invasive BCI technology. 53:14 - AI in Speech BCI: Detailed discussion on how AI is used to bridge the gap between semantic and motor signals in the brain. Addressing concerns around privacy and ensuring that BCIs only decode intentional speech, not internal thoughts.

French engineer school’s best element ❤

First

during your discussion around 54:20, you compare the initial sensations experienced by patients using your neuroprosthetic devices to those of cochlear implant recipients hearing sounds for the first time. However, auditory sensations in cochlear implants primarily involve stimulating a specific type of neurons-hair cells in the cochlea-resulting in a more direct and binary form of neural activation (either stimulated or not). In contrast, tactile sensations involve a much more complex array of mechanoreceptors and neural pathways, encoding a diverse range of stimuli like pressure, texture, temperature, and proprioception. Given this significant difference in complexity, do you believe the analogy to cochlear implants fully captures the challenges and nuances of restoring natural touch through neural interfaces? How might these complexities impact the feasibility, effectiveness, and clinical adoption of your approach compared to the relatively more straightforward neural stimulation involved in cochlear implants?

0:00 - Introduction to the BCI Award Series 0:08 - Introduction: Background and Current Role 1:03 - Career Journey: From Italy to Switzerland and the US 2:00 - Educational Background and Early Research 2:49 - First Experience with Bionic Devices and Sensory Restoration 4:41 - Transition to Studying Lower Limb Neuroprosthetics 6:08 - Motivation Behind a Focused Research Career in Neural Interfaces 8:03 - Translational Research: Bridging Technology and Patient Needs 10:00 - Patients Who Benefit from Neuroprosthetic Devices 12:28 - Why Sensory Restoration is Critical 15:52 - Challenges in Providing Sensory Feedback 19:38 - Conventional vs. Naturalistic Stimulation Approaches 27:00 - Study Aim: Improving Natural Sensation with Biometric Stimulation 29:00 - Details of the Patient Study 31:01 - Differences Between Conventional and Naturalistic Stimulation Parameters 39:46 - Measuring and Rating Sensations with Patients 48:03 - Unexpected Results and Future Research Directions 56:05 - Functional Tasks and Improved Patient Mobility 1:02:26 - Key Learnings About the Nervous System 1:12:17 - Ideal Neuroprosthesis and What’s Still Missing 1:18:32 - Comparing Neuroprosthetics with Transplant Surgery 1:23:31 - Advice for Aspiring Neurotechnology Researchers 1:28:10 - Tips for Submitting a BCI Award Proposal 1:33:30 - Final Thoughts and Future Directions

"we had g.tech and it was possible" great advertisement

"don't follow my footsteps" is the best advice!

Good to have a visual and actually see who is talking but won't be able to listen while doing dishes anymore 😅

Well done Milena😄

First

Informative

*promo sm* 🤣

Amazing and interesting podcast 🙌🏻🙌🏻🩷

🙏 *PromoSM*

🤞 Best wishes Golu