Learn Aloud: talk to any research paper, built on Vocal Bridge

A voice-first research companion that turns dense academic papers into a spoken conversation, with a tutor that explains as it highlights and an Author vs. Reviewer debate you can moderate.

The problem: reading research is too slow, too painful, too isolated

A PhD literature review can mean reading 50 to 100 papers, each one dense with notation, unfamiliar terminology, and eight-page walls of text. Santanu, who built Learn Aloud, describes the current state of reading research in three words. Too slow. Too painful. Too isolated.

It is not only students. Experienced researchers report spending more time decoding papers than doing the actual research the papers are meant to enable. The interface for absorbing new work has barely changed, a static PDF you read alone, and the cost of that shows up as hours lost per paper and ideas never reached.

There is a sharper version of this problem too. Visually impaired researchers often rely on human readers for 8 to 14 hours a day, or give up on independent research entirely. For them the static PDF is not just slow. It is a wall.

What Learn Aloud built

Learn Aloud turns a research paper into something you can question out loud instead of decode alone. The team set out to attack the real bottleneck in research, which is not getting hold of papers but understanding them. So they built a voice tutor that reads a paper alongside you and explains it in plain language, plus a second mode that pits an Author against a Reviewer so you can hear a paper argued from both sides.

What that changes is your relationship to the page. A paper stops being a static thing you push through alone and becomes something you can ask questions of, argue with, and get tested on. Santanu reports this collapses the time spent on a paper from about six hours to a 30 to 40 minute conversation, and a QR scan hands the same session off to a phone so it continues where it left off.

How Learn Aloud uses Vocal Bridge

Learn Aloud, built by Santanu Saha with Pingakshya Goswami, runs on Vocal Bridge's Voice for App surface, the pattern where the voice agent and the UI talk to each other on the same channel. When the tutor explains a section, the app scrolls to and highlights that section on screen. When the user asks for a specific reference, the agent fires a tool call to fetch and download it, and the result lands in the interface. Voice drives the view, and the view gives the voice something concrete to point at.

The architectural split is clean. Learn Aloud owns the paper parsing, the arXiv search and citation lookup agent, and the multi-persona logic that runs the Author and Reviewer. Vocal Bridge owns the voice channel, the real-time streaming, sub-second turn-taking, barge-in so you can interrupt the tutor mid-sentence, and transcript streaming.

import { useAgentActions } from "@vocalbridgeai/react";

const { onAction, sendAction } = useAgentActions();

// agent -> app: tutor is explaining a section, highlight it on screen
onAction("highlight_section", ({ page, sectionId }) => {
  scrollToSection(page, sectionId);
  setHighlight(sectionId);
});

// app -> agent: user tapped a citation, ask the agent to fetch it
const handleCitation = (refId) => sendAction("fetch_reference", { refId });

Illustrative. The actual Learn Aloud implementation handles richer paper structure and the arXiv retrieval flow.

The debate that catches gaps before reviewers do

Learn Aloud's second mode turns one paper into two voices. An Author persona presents the research, highlighting state-of-the-art results and faster training times. A Reviewer persona pushes back, asking why evaluation was limited to machine translation and how self-attention justifies capturing sequential dependency without recurrence. The Author defends, citing parallelizability on GPUs. You moderate and jump in anytime. Santanu notes this multi-dimensional view helps writers find the gaps in their own papers before a conference reviewer does.

"Imagine a world where any curious person, a student, a journalist, a researcher, an entrepreneur, can understand cutting-edge research by simply talking to it. That's the future I'm personally committed to build." Santanu Saha, creator of Learn Aloud, graduate student at Texas A&M

What's hard about voice here

What Learn Aloud pulled off is harder than it looks: a tutor that does real work mid-conversation, parsing a PDF, searching arXiv, pulling a citation, without ever leaving you in silence wondering if it heard you. A couple of things have to be right, and Vocal Bridge handles each.

• Tool calls cannot create dead air. Vocal Bridge's streaming speech and conversational fillers keep the tutor present while a lookup runs.
• You interrupt the moment a question hits. Vocal Bridge's barge-in lets you cut in instead of waiting for the explanation to finish, which is what makes it feel like a tutor rather than a recording.