Live · Real-time prosody analysis

Hear your voice.
See your emotions.

Emovo is a cross-platform system that streams your microphone audio to a hosted prosody AI model and renders the resulting emotion scores live — then archives each session for replay and long-term analytics. iOS, Android, and the web. One backend.

Scan to launch See architecture →

≤1s

visible latency

16 kHz

mono PCM-16 audio

platforms · iOS · Android · Web

48+

emotion dimensions

unified Firebase backend

What is Emovo

The gap between how you sound and how you feel — closed.

Speech prosody — pitch, intensity, rhythm, voice quality — carries a rich emotional signal that listeners decode unconsciously. Mainstream wellness apps rely on self-report (mood journals, PHQ-9), which is reflective and low-frequency by design.

One-tap capture

Tap the mic. Audio streams to Hume's prosody model and the top emotions update in real time alongside a rolling chart and a session timer.

Replay in sync

Every session saves the audio plus per-second emotion snapshots. Scrub the timeline and the chart, top emotion, and bars all re-synchronize to the playhead.

Long-term analytics

Streaks, mood-trend lines, duration-weighted top emotions, a 15-week × 7-day activity heatmap, and a peak-moments feed across all your sessions.

System Architecture

Three tiers. One data contract.

Both clients (Flutter mobile, React web) talk to a single Firebase backend for auth, data, files, hosting, and telemetry. A Firebase-Auth-protected Cloud Function provisions a per-session Hume credential, after which each client streams 1-second audio chunks directly to Hume's expression-measurement WebSocket and receives back emotion-score predictions in real time.

Clients

Mobile App

Flutter · iOS / Android

Web App

React + Vite + TypeScript

Firebase Backend

🔑

Authentication

Google sign-in

📄

Firestore

Session metadata

📦

Cloud Storage

WAV files · photos

⚡

Cloud Functions

getHumeCredentials

🌐

Hosting

HTTPS · global CDN

📈

Analytics

Usage events

☁️ Hume AI

Streaming Expression Measurement API

16 kHz mono PCM-16 1 s chunks 5 s sliding window

Clients ⇄ Firebase — sign-in, sessions, files, hosting, telemetry

Cloud Functions ⇢ Clients — Hume credential to authenticated callers only

Clients ⇄ Hume — direct WebSocket; audio up, emotion scores down

Tech Stack

Built with the tools we love.

A deliberate, narrow set of technologies — chosen so the same data contract serves mobile and web with no per-platform plumbing.

Flutter

Dart

React 18

TypeScript

Vite

MUI v6

Firebase

Cloud Functions

Node 22

☁️Hume AI

🐻Zustand

🟣Riverpod

iOS

Android

Features

Designed for daily use.

Every feature lives on every platform. Sessions saved on the phone replay seamlessly on the web — and vice-versa.

Live capture

Gradient card for the dominant emotion with a delta arrow. A rolling 3-emotion line chart. Animated top-5 bars. A clean session timer. One-tap to start, one-tap to stop.

Web Audio API + AudioWorklet (web)
record package (mobile)
Identical byte streams cross-platform

Sessions, replayable

Every completed session is stored with its full audio and per-second emotion scores. Tap a peak moment to jump straight to it. The chart, top-emotion card, and right-now bars all re-synchronize to the playhead.

WAV stored in Cloud Storage
Snapshots tagged with PCM-byte offset (no clock drift)
Same Firestore doc on every device

Analytics that respect your time

Streak, session count, total minutes. Mood-trend line chart over the three dominant emotions. Duration-weighted top-emotions bars. A 15-week × 7-day activity heatmap. A cross-session peak feed.

7-day · 30-day · all-time ranges
Hand-rolled charts — identical pixel-for-pixel
Firestore onSnapshot · push updates

Data flow

How a single second becomes a feeling.

1

Mic → PCM

Audio is captured natively, resampled to 16 kHz mono Int16 PCM inside the audio thread.
2

Chunk + WebSocket

Each 1-second chunk is wrapped in a WAV header, base-64 encoded and sent over a WebSocket with a payload_id.
3

5 s sliding window

Hume analyzes each chunk against a 5-second context so prosody continuity isn't lost between chunks.
4

Live render

Predictions arrive asynchronously and feed the rolling chart, top-emotion gradient, and animated bars at 60 fps.
5

Persist

On stop, the full PCM is wrapped into a single WAV, uploaded to Cloud Storage, and a session document is written to Firestore.

Engineering challenges

The hard bits — and how we solved them.

iOS / Safari

`AudioContext` starts suspended

Defensive audioCtx.resume() after the user-gesture so the worklet never silently no-ops.

Empty predictions on isolated chunks

Use Hume's 5 s sliding window so each 1 s chunk is analyzed against continuous voiced context.

Replay sync

Wall-clock drift desyncs charts

Tag every snapshot with the cumulative PCM byte count → exact audio-clock offset for playback sync.

Cross-platform

Two codebases drifting apart

Single source-of-truth Firestore schema; mirrored module structure; same Material-3 seed colour #6750A4.

Results

What we shipped.

End-to-end realtime emotion streaming with sub-second visible latency on every platform — and full feature parity across iOS, Android, and web.

< 1s

Visible end-to-end latency

From mic capture to rendered emotion score, on both clients.

100%

Cross-platform parity

Every screen, analytics widget, and theme state present on web and mobile.

Firebase project

Hosting, Functions, Auth, Firestore, Storage — no separate infra.

⇄

Fully interoperable sessions

A session saved on the phone replays on the web — same Firestore doc, same WAV.

Future Work

Where Emovo goes next.

📊

Longer-horizon analytics

Weekly / monthly aggregates, day-of-week effects, calendar overlays.

💬

Empathic Voice Interface

Integrate Hume's conversational EVI endpoint for two-way voice interaction.

📱

On-device inference

Lightweight prosody models for offline / privacy-preserving capture.

⚙️

PWA & background sessions

Native PWA install on web; background-recording on mobile.

🧪

User study

Does passive voice tracking improve emotional self-awareness vs. mood-journal baselines?

Try it

Scan, speak, see.

Two builds, one experience. Pick the device in your hand and start a session in under five seconds.

Web App

React + Vite · Firebase Hosting

Drop web.png
into assets/qr/

emovo-e06a8.web.app