Voice Mode

How it works

What you get

A voice interface worth using.

Natural

Hold a real conversation

Not a voice assistant with trigger phrases. Aria speaks at the cadence of a thoughtful consultant, with pauses where pauses belong.

Persona

Calm, measured voice

ElevenLabs voice tuned for consulting: stability 0.55, style 0.15. No over-dramatization. No uptalk.

Continuity

Same conversation, any mode

Switch between typing and speaking mid-thread. The transcript preserves everything. Aria doesn't lose context across modalities.

Interrupts

Talk over Aria

Interrupt the response whenever. Aria stops cleanly, listens, and continues from where your correction lands. Natural conversation flow.

Privacy

No audio stored

Speech is transcribed on your device. Only the text reaches our servers. Transcripts can be reviewed and redacted before they're saved.

Fallback

Text on unsupported browsers

Safari without the flag? Firefox? The page detects it and shows a typed-input fallback — same Aria, same answers, no degraded experience.

Performance

Measured in milliseconds.

Voice has a latency budget. Exceed it and it feels broken. These are our targets — and our actuals.

<500ms

First audio

Time from end-of-speech to the first word of Aria's response. Streaming synthesis, not wait-then-play.

98%

Transcription accuracy

On US English, measured against professional transcription. Graceful degradation on accents and noise.

12

Languages

Full support for the same 12 languages as stakeholder interviews. Native speakers tuned each voice.

2 min

Average session

Long enough to work through a question, short enough to keep voice conversational. Most sessions are ~4 turns.

The <500ms response feels like Aria was already thinking about my question. I stopped asking Aria things via text. Voice is just how we work now.

Margaret Chen

Chief Operating Officer · Vantage Industrial Group

Specs

Under the hood.

Speech-to-text

Web Speech API (webkitSpeechRecognition) for supported browsers. Transcription on-device; only the final text is transmitted.

Text-to-speech

ElevenLabs Turbo v2.5 via /api/tts proxy. API key stays server-side. Streaming MP3 begins within ~300ms of the request.

Voice settings

Stability 0.55 · Similarity boost 0.8 · Style 0.15. Calibrated for a calm, measured consulting persona — no dramatic inflection.

Browser support

Chrome, Edge, Safari (with speech flag). Firefox and unsupported browsers receive an automatic text-input fallback with no degraded experience.

Privacy

Audio is never recorded, transmitted, or stored. Only the transcript reaches our servers, and it follows the same redaction rules as chat.

Interrupts

Tap the orb mid-response to interrupt Aria. Response audio stops within 80ms. The interrupt point is preserved in the transcript.

Try it now

Speak with a live Aria.

This is real. Click the orb, allow microphone access, and ask Aria anything about the Vantage Industrial assessment. Aria's answer is streamed back in natural voice.

This session uses the live ElevenLabs TTS and your browser's Web Speech API. Nothing is stored. The demo response is scripted — in production, Aria answers with full context from your assessment.

FAQ

Common questions.

Explore

More of Aria.

Each capability works standalone — or together as a full operational intelligence suite.

01

Ready to speak with Aria?

Voice Mode is included on every Aria engagement. Start a conversation in chat, then switch to voice whenever it feels more natural.

Try the full demo See all capabilities→

Speak with Aria.

Speak naturally

Browser-native STT

Full context, same Aria

Natural voice in <500ms