Linkubit
GuidePublished 19 September 2026 · 5 min read

Hindi voice notes on WhatsApp Business: transcribing them, replying in the same language, and the normalisation bug that breaks STOP

Voice notes are the most detailed and the last-answered messages a business gets. Transcribe on arrival, keep the transcript on the message, reply in the language spoken, extract the facts, and never strip Devanagari's vowel signs.

0:09
Transcript · Hindi
auto
“दो पैलेट मुंबई से चेन्नई, शुक्रवार तक, डोर टू डोर। रेट बताओ।”
दो पैलेट BOM → MAA, डोर टू डोर, शुक्रवार तक: ₹12,500। बुकिंग कन्फ़र्म करूँ?Agent
In short

A large share of enquiries from Indian customers arrive as voice notes, often in Hindi or a mix of Hindi and English. A business that reads them an hour later loses the lead; one whose assistant transcribes the note, understands it and replies in the same language within a minute keeps it. The transcript belongs on the message, where the team and the next turn can read it, and the reply should be text — short, in the customer's language — unless they asked for a call.

On this page
  1. Why voice notes matter more than they look
  2. What good handling looks like
  3. The Hindi problem in most tools
  4. Languages beyond Hindi

Why voice notes matter more than they look

A voice note is the message a customer sends when typing is slower than talking — walking, driving, on a site, or in a language their keyboard does not have. It is often the most detailed message in the thread: the whole requirement in forty seconds. And it is the message most businesses answer last, because someone has to listen to it. On a shared number with three people, a voice note in Hindi at 7 pm is answered at 10 am, if the person who speaks Hindi is in.

What good handling looks like

  1. Transcribe on arrival. The note becomes text within seconds, in the language it was spoken, with Hinglish kept as Hinglish rather than forced into one script.
  2. Keep the transcript on the message. In the thread, on the contact's timeline, and in what the assistant reads on the next turn — a voice note with no text is dropped from the conversation's history for having no body, and the assistant answers as if it never arrived.
  3. Answer in the same language, as text. A Hindi note gets a Hindi reply. Text, because it is read at once and searchable later; a voice reply is for when the customer asked for one.
  4. Extract the facts. Budget, locality, dates — whatever the note contained — written onto the lead the same as a typed answer.
  5. Hand over on the same rules. A complaint spoken is a complaint; the handover rules read the transcript.
0:41
Saw your 2BHK ad in Whitefield. Price?
₹82 lakh, ready to move. Are you looking to buy in the next three months, or later?Agent
Next month. Can we see it Saturday?
Replied in 41 seconds · site visit booked · lead qualified
A voice note answered in a minute is a conversation; one waiting for someone to listen is a lead going cold.

The Hindi problem in most tools

Two things go wrong. Transcription trained on English mangles Hindi and gives up on Hinglish. And text handling that "normalises" input by stripping accents destroys Devanagari — the vowel signs in बंद करो are combining marks, and the usual normalisation folds them away, so a STOP in Hindi matches nothing and the customer keeps getting messages. A tool that handles Hindi has been tested on Hindi, not on transliterated English.

In Linkubit

An inbound voice note is transcribed by the AI model, the transcript and its language are written onto the message, and the assistant replies in that language as text. STOP, बंद करो, and their equivalents in the other supported languages all suppress the person, because the matching folds Latin accents only and leaves Indic scripts whole. A voice-note minute costs half a credit.

Languages beyond Hindi

The same handling applies to Tamil, Telugu, Kannada, Marathi, Bengali, Gujarati and the rest: transcribe in the spoken language, reply in it, match the opt-out in it. The assistant's persona and the templates are written once and rendered in the customer's language; a business does not maintain six copies of its playbook.

Can WhatsApp Business transcribe voice notes?

The WhatsApp app itself offers on-device transcription in some languages for the reader; it does not give a business the text to act on. On the Business Platform the audio arrives as a file and the business's software transcribes it — that is what Linkubit does on arrival.

Should the reply to a voice note be a voice note?

Usually not. A text reply is read immediately, quoted back later, and searchable. Send voice when the customer asked for it or when the answer is genuinely easier spoken.

What does transcription cost?

In Linkubit, half a credit per minute of audio. Meta charges nothing extra for receiving audio.

Two pallets Mumbai → Chennai, door to door, by Friday: ₹12,500 from the rate card. Shall I send the payment request?Agent
Yes, go ahead.
Order · Sai Traders
Freight BOM → MAA · 2 pallets₹12,500
Pay with UPI
to your account · never ours
Paid ₹12,500 by UPI · deal won · attributed to Meta lead form
The end of the same conversation: a quote from the rate card and a payment request, in the customer's language.

Read next

Try it on your own number. Linkubit answers, qualifies and quotes on WhatsApp from your catalogue, and adds nothing to Meta’s per-message price. Start free — fourteen days of Pro, no card — or read the pricing.