Earnings Call Analyst: voice-driven earnings call analysis for investors, powered by Vocal Bridge

A voice agent that ingests earnings call recordings, structures them into highlights, tone, and Q&A behavior, and answers transcript-grounded questions out loud.

The problem

An earnings call is an hour of dense, fast-moving signal. Inside it sit the numbers that move a stock, the topics management chose to repeat, the tone behind the guidance, and the questions analysts pressed hardest on. Most of that is buried in a long recording or a wall-of-text transcript that nobody has time to scrub through during a busy reporting season.

The usual tools force a tradeoff. A summary is fast but loses the exact wording, and exact wording means scrolling through the full transcript with a timestamp in your head. Neither answers the question an analyst actually has in the moment, which is some mix of "what mattered here" and "show me precisely what was said."

What they built

Earnings Call Analyst lets an investor interrogate an earnings call by voice. Ask what stood out and you get a structured read back covering the focus areas, a read on management tone, and analyst activity. Ask for an exact passage and the agent retrieves it word-for-word and pins it to its timestamp in the call. Two very different jobs, summary and verbatim retrieval, in the same conversation.

The point is that an analyst can interrogate an hour of audio the way they would ask a colleague who had the transcript open. Voice fits because analysts ask their follow-ups out loud, and a conversation beats scrubbing a full transcript for the one line that matters.

How they use Vocal Bridge

Earnings Call Analyst is built on Vocal Bridge's Voice for App surface, where the voice agent and the app UI exchange actions over the same real-time channel. The agent runs a query against the indexed call, and the dashboard reflects status while the work happens. When the answer is ready, the voice channel delivers it and the agent offers the next follow-up.

The architectural split is clean. The team owns the financial intelligence: ingesting recordings, extracting highlights and topics, segmenting prepared remarks from Q&A, scoring management responses for directness, and retrieving transcript spans with timestamps so every answer is traceable. Vocal Bridge owns the voice: the real-time channel, streaming-aware text-to-speech, barge-in, transcript streaming, and the conversational fillers that keep a user oriented while a query runs.

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

const { onAction, sendAction } = useAgentActions();

// agent -> app: a grounded answer arrives with its transcript span
onAction("transcript_segment", ({ text, startMs, endMs }) => {
  setSegment({ text, start: fmt(startMs), end: fmt(endMs) });
});

// app -> agent: user picks a call to analyze
const handleSelectCall = (callId) => sendAction("load_call", { callId });

Illustrative. The actual implementation handles richer earnings entities and scoring.

What's hard about voice here

This job lives or dies on grounding, and voice adds a second bar on top.

• Answers have to come from the transcript, not memory. A confident answer that drifts from the call is worse than none, so the agent retrieves real spans and cites timestamps rather than generating. (The team reports zero hallucination as their own claim.)
• Retrieval takes a beat. Pulling and reading a verbatim passage is not instant. Vocal Bridge's conversational fillers cover that gap so the user stays grounded instead of waiting on silence.

That combination is the difference between a summary you can quote and one you have to double-check.

"Real-time, transcript-grounded voice queries with zero hallucination." Muthu Sankar, Earnings Call Analyst