Product features

From spoken answers to structured responses.

See how AmphiVox understands natural speech, validates each value, and keeps creators in control from form building through final delivery.

Try a live form Create a voice form

How it flows

Speech to response

Live flow

Spoken answer

“Two units tomorrow, COD.”

Structured response

Three fields updated together

Publish

Link, widget, or React

Deliver

AmphiVox, Sheets, or Excel

01 - Conversation becomes structure

Let people answer naturally. Keep the form structured.

One spoken answer can fill several fields. Later corrections update the same ledger, so respondents are never trapped in a rigid prompt-per-input exchange.

Transcript - turn 1

"The Friday workshop — make it two guests."

FieldValueHeard asStatus

Field

Workshop date

Value

Friday

Heard as

Friday workshop

Status

Collected

Field

Guest count

Value

Heard as

two guests

Status

Collected

Transcript - turn 2

"Actually, make that three."

Field

Guest count

Value

Heard as

make that three

Status

Corrected

01 / 04

Multi-field capture

It cuts both ways: the assistant groups related questions into one natural ask, and whatever a respondent volunteers on top of that is mapped in the same turn — however many fields it happens to cover.

"Nandini Layout, Bangalore" -> Street + City, from one answer

02 / 04

Corrections

A later turn replaces a value the respondent already gave, and the form on screen updates with it. Nothing has to be restarted to fix one answer.

03 / 04

Hinglish, and 45+ languages

Across the Indian languages, people mix — Hinglish, Tanglish, a Kannada sentence with English nouns in it — and the assistant follows the mix instead of forcing one language. On the worldwide stack, answering in your own language works, and so does falling back to English mid-sentence when a word is easier that way.

Delivery kal chahiye -> Delivery date: Tomorrow

04 / 04

Already-answered fields

When the form already carries values — a draft the respondent left, an earlier session, or values the host application passes in — those fields arrive answered. The assistant does not re-ask them; it starts from what is still missing.

Fields marked PII, and manual-only file, URL, and legal fields, are never prefilled through voice.

01 / 04

02 - Validation and sensitive boundaries

Validate the details that usually break forms.

A postal code fills in the city and state, and whatever is still missing gets asked for. Phone numbers are checked against country context and stored in international format. Sensitive manual-only fields stay on the form without the voice agent ever asking for them.

Address lane

A postal code fills in city and state

Spoken560034, Koramangala

CityBengaluruDerived

StateKarnatakaDerived

CountryIndia

Still neededHouse or building number

Assistant: "Got Koramangala, Bengaluru — what is the house or building number?"

Derived values are read back, not locked: a respondent who says a different city keeps theirs, and the assistant flags the mismatch once. A postal code that matches nothing is re-asked rather than stored.

Phone lane

Spoken digits become E.164

Spokenplus nine one nine eight...

RegionIndia

Stored+91 98765 43210

Too few digitsRe-asked

Digits are taken from the transcript rather than the model's retelling of it, so a run of repeated numbers survives the trip. A spoken country code wins over the field's default region.

Manual-only lane

The voice agent never asks for it

Aadhaar number PII

Voice pathSkipped

Manual form pathEnabled

Delivery pathNormal submission

Entered manually, the value does not enter the voice-agent LLM context. PII marking controls the designed collection path; it cannot redact something a respondent chooses to say aloud while the microphone is active.

03 - Creator workbench

Turn an existing form into an editable voice form.

Import a screenshot or PDF, review the extracted fields and notes, then edit the form before previewing the respondent experience.

Uploaded source

Scholarship form.pdf

Import screenshots or PDFs, then review and edit the extracted form before anything is published.

Extracted structure

Title and description

11 fields detected

Address and phone fields typed

Consent field marked legal

Import note: handwriting on page 2 could not be read. Review before publishing.

Editable form

Full name

Text

Phone number

Phone

Validated

Residential address

Address

Resolved

Aadhaar number

PII

Manual-only

Upload marksheet

File

Drive required

Add, remove, reorder, and configure fields directly. Preview runs the form and voice assistant together before publishing.

04 - Publish surface

Share a hosted form, or bring voice to your existing page.

Use an AmphiVox link, a JavaScript widget, or a React integration depending on how much control your site needs.

Direct form links

Publish and share the URL. Nothing to install on either side — the respondent opens it in a browser and starts talking.

Published form

https://amphivox.com/form/k3n8v2qh7d

Each published form gets its own generated link.

Fastest path

Campaigns, surveys, applications — anywhere you can paste a link.

Public or private

Keep it open, or require sign-in and an allowed-email list.

Single response

With sign-in required, each respondent can submit once.

05 - Delivery map

Route structured responses where your team works.

Keep submissions in AmphiVox, write them to Google Sheets, send them to Microsoft Excel, and store uploaded files in Google Drive.

Structured submission

One field schema, the destination you choose.

Route responses to AmphiVox, Google Sheets, or Microsoft Excel. Uploaded files continue to the connected Google Drive folder.

AmphiVox storage

Encrypted stored responses your team can review and export inside AmphiVox.

Google Sheets or Microsoft Excel

Send each structured submission to a connected Google Sheet or Microsoft Excel workbook.

Google Drive

Files collected by file fields continue to the creator's connected Google Drive folder.

06 - Trust ledger

Be clear about what the assistant handles.

Respondents review the collected fields before submission. Raw voice recordings are not saved, while diagnostic text logs and destination-specific retention are explained plainly.

Raw audioRecordings of a respondent's voice are not saved. Audio is streamed for speech recognition, converted to text, and discarded as the session runs.

Conversation textDiagnostic logs can contain conversation text and collected answers for seven days.

SubmissionsSubmitted-response retention depends on the creator's selected destination.

Respondent reviewCollected fields are visible and editable before submit.

AI assistantRespondents are told they are speaking with an AI voice assistant before the voice session begins.

Read the Privacy Policy

Ready when you are

See it working in a real form.

Try a live demo Create a voice form Still deciding? Read the FAQs