.png&w=256&q=75)
🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
ClearVoice 2.0
Over 70 million individuals worldwide live with speech dysfluencies such as stuttering, blocks, repetitions, and mid-sentence restarts. In our voice-first digital world, conventional AI systems fail them: traditional Voice Activity Detection (VAD) abruptly terminates recording during brief pauses, speech-to-text creates garbled repetition bloat, and unconstrained language models introduce dangerous factual hallucinations like altering numbers or stripping negations. ClearVoice 2.0 solves this by introducing a real-time, sub-second speech repair and neural resynthesis pipeline. Built with a high-performance polyglot architecture, our Python core engine uses PyTorch Silero VAD with a custom 1,500ms adaptive silence window and 512-sample frame slicing to eliminate premature cutoffs, streaming live audio directly to AssemblyAI's Realtime WebSocket API. When speech is transcribed, our Node.js intelligence server leverages Groq-hosted open-weights models alongside a 6-utterance rolling context buffer to intelligently remove dysfluencies while preserving intentional emphasis. To guarantee zero hallucination, a deterministic Meaning Validator rigorously inspects every repair, ensuring critical numbers, dates, negations, and named entities remain 100% faithful to the original speech. If any drift is detected, an automated zero-trust fallback engine serves a conservatively cleaned or raw transcript. Approved speech is resynthesized through a resilient 3-tier failover engine: primary high-fidelity streaming via ElevenLabs PCM16, secondary on-server system TTS via pyttsx3, and tertiary browser-level client synthesis. Deployed in a Dockerized AWS EC2 backend with an interactive Next.js 16 frontend on Vercel, ClearVoice 2.0 democratizes voice interfaces and empowers dysfluent speakers with clarity, dignity, and confidence.
30 Sep 2026
Echo Canvas
Echo Canvas is an accessible AI art platform designed to make creative expression available to everyone, including people with disabilities that affect their ability to use traditional tools. We recognize that not everyone can articulate their vision through precise text or steady hands, which is why we've built a platform that starts with inclusivity at its core. Our current platform empowers users through two primary, accessible input methods: natural language text and voice. A user can either type a description or simply speak their vision aloud, whether it's a detailed scene or an emotional sound. Echo Canvas's intelligent backend then analyzes this input, detecting not just the content but also the underlying emotional tone and any specific accessibility needs mentioned, such as color vision deficiencies. The system then generates a beautiful, optimized piece of art. For users who are blind or have low vision, we complete the creative loop by providing an AI-generated, rich audio description of the artwork, allowing them to fully experience their creation. By focusing on voice and text, Echo Canvas bridges the gap between imagination and artistic creation, proving that the power of art should be, and now can be, accessible to all.
19 Nov 2025
