The Future of Human Communication: AI, Bionics and Voice Technology | Konrad Zielinski x Shekhar Natarajan

Overview

This episode of Tomorrow Today features Conrad, an entrepreneur and researcher who lost his biological voice after treatment for laryngeal cancer. Drawing on his experience in speech technology, he developed a non-invasive electronic system that amplifies and reshapes electrolarynx speech, helping users communicate more clearly and expressively.

The conversation moves beyond speech intelligibility to a harder problem: preserving the timing, pitch, emotion, laughter, and other vocal signals that make communication feel human. Conrad argues that effective voice augmentation requires an appropriate division of labor—the user controls intention and emotion, while AI helps preserve a recognizable personal voice.

The discussion also examines technical and commercial challenges, including real-time latency, on-device inference, pricing, reimbursement, voice-data ownership, and authentication. It closes with a broader vision in which assistive voice devices become mainstream wearables, reducing the stigma experienced by people who depend on them.

Segment Summary

[01:13] Introducing AI-Powered Bionic Speech

[Fact] The episode is recorded at TechCrunch Disrupt and introduces Conrad as a builder working at the intersection of bionics and AI.

[Fact] Conrad uses an electronic device during the interview to make his voice louder and easier to understand.

[Fact] He lost his biological larynx after being diagnosed with laryngeal cancer and having his vocal folds removed.

[01:54] From Personal Diagnosis to a Non-Invasive Product

[Fact] Conventional options can produce a quiet, robotic voice or require an invasive procedure involving the throat.

[Fact] Conrad’s team pursued a non-invasive alternative based on electronics, without additional throat surgery.

[Fact] He says the technology enables him to run a business, complete a PhD, maintain an active social life, and even perform at a concert.

[04:51] Target Population and Origins of the Company

[Fact] Conrad says approximately five million people worldwide have laryngeal cancer and may require a laryngectomy; this population is the company’s initial market.

[Fact] The company began as a mission-driven project rooted in Conrad’s own needs before expanding to additional users and conditions.

[Fact] Conrad had already worked with speech and communication technology before his first diagnosis in 2018, so he recorded his voice and assembled collaborators early in the process.

[06:27] Availability, Users, and Product Roadmap

[Fact] The company began individual sales of an earlier-generation device that amplifies electrolarynx speech and enhances selected frequencies.

[Fact] Conrad says more than 600 people worldwide use earlier components of the technology, initially distributed through Polish and American partners.

[Fact] A newly developed generation was demonstrated during the interview and is planned for sale in 2027.

[Fact] The company is expanding from distributor-led sales toward a direct-to-consumer model.

[07:35] How Vocal Expression Shapes Thought

[Fact] Conrad explains that changes in pitch and frequency communicate emotions such as anger, love, and affection.

[Fact] He recalls that devices with flat intonation limited his ability to express certain ideas and even made those ideas harder to think through.

[Fact] Gaining manual control over intonation helped him articulate thoughts that had previously remained difficult to express.

[08:41] A Non-Invasive Alternative to Brain Implants

[Fact] Conrad’s roadmap includes systems that interact with brain-related, muscular, and cardiac signals, but the company favors non-invasive methods.

[Fact] He contrasts this approach with implanted brain-computer interfaces, emphasizing accessibility to the general population.

[Fact] A future device using additional biological signals is planned for 2028.

[09:59] Personal AI and Emotion-Aware Communication

[Fact] Conrad describes a shift from text-only AI toward systems that continuously listen and respond rather than waiting for discrete prompts.

[Fact] He expects personal language models to become integrated into wearable devices and assist with tasks such as low-delay translation.

[Fact] He identifies a major challenge in preserving emotion and timing when AI mediates conversations between people from different languages and cultures.

[Inference] The company’s long-term ambition extends beyond restoring speech and toward creating a general interface between human intention, AI assistance, and social interaction.

[11:43] Why Words Alone Are Not Enough

[Fact] Conrad says rapid participation is essential in conversation, citing a user who avoids the theater partly because he cannot engage quickly and his device attracts attention.

[Fact] He calls himself a “scientist of unwords,” arguing that laughter, crying, shouting, affection, and similar vocalizations are harder to reproduce than ordinary words.

[Fact] The product therefore needs to solve three connected problems: the sound of the voice, user control over it, and conversational timing.

[13:16] Latency as the Central Technical Barrier

[Fact] Conrad says some voice-conversion systems have delays of roughly 300 to 1,000 milliseconds, while natural interaction requires latency closer to 50 milliseconds.

[Fact] Current high-quality speech-to-speech systems can operate at approximately 200–300 milliseconds and avoid the older speech-to-text-to-speech pipeline.

[Fact] Approaches reaching roughly 70–80 milliseconds exist, but their output quality remains insufficient.

[Inference] Conrad expects embedded 50-millisecond conversion within two or three years, contingent on better chips, smaller and faster models, and more training data.

[15:38] Constraining AI with Acoustics and Biomechanics

[Fact] The team includes musicians and audio engineers and uses knowledge from acoustics, speech therapy, medicine, and biomechanics.

[Fact] Rather than asking a model to infer everything from large datasets, the team constrains it with known physical and acoustic properties of speech.

[Fact] Conrad says modern voice models are generally language-agnostic or can be fine-tuned relatively easily for additional languages.

[17:04] The Next Breakthrough: Seamless Wearables

[Fact] Conrad identifies the next major step as embedding speech technology into wearable devices.

[Fact] He describes three necessary components: seamless low-latency operation, awareness of human-to-human conversational context, and AI agents working in the background.

[Fact] He notes that unconstrained live AI systems can unexpectedly switch languages, while explicit constraints can keep them operating in a selected language.

[19:13] Pricing and Reimbursement

[Fact] Conrad says the device he is using costs about $1,500, an earlier generation costs about $515, and a future version for professionals and the broader public may cost about $3,000.

[Fact] The lower-cost model was priced to fit Poland’s reimbursement limit, allowing eligible users to receive it without paying the full cost themselves.

[Fact] The company intends to adapt pricing to reimbursement limits in markets such as Germany, with users paying extra for higher-end models if desired.

[20:39] Learning Directly from Users

[Fact] Conrad says about 200 customers purchased directly from the company, while a wider group uses devices obtained through distributors.

[Fact] He offers short online consultations to understand individual speech problems, including cases the current product cannot yet solve.

[Fact] When another technology, such as a brain-computer interface, may be more appropriate, he is willing to refer the person to another specialist.

[22:14] Balancing Founder Experience with Broader Needs

[Fact] Conrad is both the company’s CEO and a user of its product.

[Fact] He acknowledges that his needs do not represent the entire user population because he is a comparatively young user and his current device already works well for him.

[Fact] Product decisions therefore prioritize the broader population rather than optimizing only for Conrad’s personal preferences.

[23:00] Voice Data Ownership and Consent

[Fact] Conrad considers secure storage especially important because recorded voices may later be recreated by AI.

[Fact] He argues that users should own their voice data and remain free to move it to another provider.

[Fact] His position was shaped by difficulty retrieving his own recordings from a previous provider after recording his voice before surgery.

[Fact] Although more shared data improves model training, he believes participation must remain voluntary because voice recordings may expose highly personal conversations.

[25:28] Edge Computing and Human–Machine Roles

[Fact] Conrad says real-time inference should run on the device, while model training can happen elsewhere before a distilled model is deployed locally.

[Fact] In his current system, he manually controls emotional expression while the device controls aspects of how his recognizable voice sounds.

[Fact] He describes this separation as a division of labor between human and machine: the person supplies intent and emotion, while AI preserves vocal identity.

[26:53] A Cyborg Who Remains Human

[Fact] Asked whether AI augmentation makes him an agent or a human, Conrad says he considers himself one of the first cyborgs.

[Fact] He immediately emphasizes that he is still human despite relying on augmented speech.

[Inference] His answer frames bionic technology as an extension of human agency rather than a replacement for personal identity.

[27:42] Making Assistive Technology Mainstream

[Fact] The technology could eventually serve people with Parkinson’s disease, dysarthria, cerebral palsy, and other brain-related injuries.

[Fact] The company nevertheless plans to expand toward the general population as well as people who have undergone laryngectomy.

[Fact] Conrad argues that assistive devices will lose their stigma when people with healthy voices also choose to use them, much as glasses can function as both aids and ordinary accessories.

[Fact] He wants users with disabilities to be seen as people choosing technology, rather than being defined by disease or necessity.

[29:05] Authentication in a World of Synthetic Voices

[Fact] The host asks how a voice-authentication system can distinguish a real person using an augmented voice from an autonomous AI agent.

[Fact] Conrad predicts that conventional voice authentication will cease to make sense within two or three years because AI will be able to generate arbitrary voices.

[Fact] Existing systems can mark audio as AI-generated, but Conrad is not aware of a system that also identifies the speaker as a real person using synthetic speech because of a disability.

[Fact] He proposes combining synthetic-voice disclosure with a secure identity token, voiceprint, or comparable credential tied to the individual.

[Inference] Future authentication will need to verify identity independently of whether the audible voice is biological or generated.

[33:02] Closing Perspective

[Fact] The host praises the project as an AI use case that directly improves human life rather than merely automating routine content tasks.

[Fact] Conrad says the questions reflect problems his team is currently confronting and expects the company to work on them over the next four years.

Podcast Commentary/Summary

The episode’s strongest contribution is its treatment of voice as more than intelligible words. By focusing on laughter, timing, pitch, affection, and conversational spontaneity, it shows why restoring communication is a deeper engineering challenge than producing understandable synthetic speech.

Conrad’s dual perspective as founder and product user gives the discussion unusual practical depth. The conversation connects personal experience with concrete questions about latency, hardware constraints, reimbursement, privacy, consent, and inclusive product design.

[Inference] Some market-size figures, product timelines, prices, and performance forecasts are presented conversationally and are not independently substantiated within the transcript. The episode is therefore most valuable as a founder’s technical and ethical perspective rather than as a verified market report.

[Inference] This episode is particularly suitable for listeners interested in assistive technology, speech AI, wearable computing, accessibility, product strategy, and the social implications of human augmentation.