Blog · WhatsApp and channels
Is there an AI that transcribes your customers' WhatsApp voice notes automatically?
Yes: Meteor transcribes the voice notes customers send to your business WhatsApp as they arrive. What your team sees, what the agent does and the limits.
Is there an AI tool that transcribes WhatsApp voice notes from customers automatically?
Yes. Meteor transcribes every voice note a customer sends to your business WhatsApp number as it arrives, when the number is connected through the official API. Nobody forwards the audio: the text appears under it in the inbox, the AI agent reads it before replying, and your systems get it through the API. It uses Energy from your balance, and today the expected language is Spanish.
Yes, and you do not have to forward anything to another tool. When your company’s WhatsApp number is connected to Meteor through the official API, every voice note a customer sends you is turned into text on arrival, as long as you have Energy left. Your team reads that text in the inbox, the agent uses it to reply, and your own system gets it through the API.
Two searches look alike here. If you want to read the voice notes in your own chats, the WhatsApp app already transcribes them on the phone, from Settings, Chats. That transcript stays on that phone. If you run a business and want each customer’s voice note written down in the conversation, for the whole team and for the agent, that is what Meteor does.
AI tools that transcribe WhatsApp voice notes, and which ones are automatic
There are four kinds of tools, and only some of them work without anyone doing anything. Reviewed in November 2026 on each vendor’s page:
| Tool | How it works | Automatic for your customers’ voice notes? |
|---|---|---|
| WhatsApp’s own transcripts | You turn them on in Settings, Chats; the text is generated on your phone (WhatsApp) | Only for whoever has the phone in hand; your team and your systems do not see it |
| A WhatsApp Web extension, such as Kaption | Adds a transcribe button to each voice note in WhatsApp Web, in Chrome or Edge (Kaption) | On the computer where it is installed |
| A transcription contact, such as Transcribbit or VozClara | You forward the voice note to their WhatsApp number and the text comes back (Transcribbit, VozClara) | No: someone forwards each voice note |
| A platform on the official API, such as Meteor | Every voice note that reaches your business number is transcribed on arrival | Yes: in the conversation, for the whole team, the agent and the API |
If you only want to read the voice notes in your own chats, the first three are enough. If you want every customer voice note in writing without anyone touching it, the number has to be on the WhatsApp Business API and a platform has to do it on arrival.
What happens to a voice note when it arrives
| Step | What happens | What you see |
|---|---|---|
| 1. The customer sends the voice note | Meteor downloads the file and stores it in private storage, attached to the contact | The audio in the conversation |
| 2. The balance is checked | Transcription uses Energy; with no balance, the audio stays without text | The audio, with no text under it |
| 3. The file is prepared | If the format is not one the transcription engine takes as is, like the AAC some phones send, AMR or 3GP, it is converted to OGG first | Nothing: it is automatic |
| 4. The transcription engine turns it into text | The text is saved next to the audio | “Transcribing…” while it runs |
| 5. The text joins the context | The Met receives the audio with its transcript before it prepares a reply | The Met’s reply, if one is assigned |
| 6. It goes out to your systems | The text travels in attachments[].transcript through the API and webhooks | Whatever your integration does with it |
If anything fails along the way, the message is saved anyway. Transcription is an extra on top of the message, never a condition for receiving it.
What your team sees in the inbox
In the conversation inbox every voice note shows its player, and the text appears right below it. Skimming the text is faster than pressing play. When the note is long, roughly 40 words or more, the text folds so it does not take over the conversation, and opens with one click.
The text stays in the contact’s history. Whoever picks up the case tomorrow reads what the customer said without listening to the audio again.
In the conversation settings you choose the AI level for voice notes: Ágil, Brillante or Genio. The product describes Brillante as better with background noise and accents, and Ágil as faster and cheaper.
What the agent does with the text
If the channel has a Met assigned, the Met receives the transcript as part of the message and replies to what the customer said, as if it had been typed. If the voice note includes an order number, a date or an address, the Met has them as text to look up or confirm with the tools you connected.
With no Met, your team replies from the same inbox with the text in view.
Replying with a voice is a separate capability. If the Met has a voice set up, it answers in a voice note to people who spoke to it in audio. You turn that on or off in the Met settings, and you can also have a share of the replies to written messages go out as voice. If the voice fails, the customer gets the written reply.
Your team’s voice notes too
Voice notes an agent records in the inbox are transcribed just like the customer’s, and so are the ones the owner sends from the phone app when the number is connected with coexistence. They use Energy like incoming ones. The Met’s replies are not transcribed: they are already text.
For developers: where the text lives
Through Meteor’s public API, every message with audio carries the text in its attachment:
"attachments": [
{ "type": "audio", "url": "https://…/nota.ogg", "transcript": "Hola, necesito una guía para mañana" }
]
- The
contentfield still says"[Audio]", so an integration that already spots audio by that text does not change. - The
contact.message.receivedwebhook includestranscriptwhen the transcription was ready within the first 30 seconds. If it took a little longer, the text usually shows up when you fetch the contact’s messages through the API. contact.message.sentcarries the same field for your team’s voice notes.
The details are in the webhooks guide on the developer portal, linked in the sources.
What it costs
Each transcribed voice note is deducted from your Energy balance based on its length. On the Estudio to Campus plans it uses less than one US cent per minute. The same voice note is never charged twice: if it already has text, that text is reused. The Estudio plan costs US$99 a month with US$10 of Energy included; the rest is on Pricing. Meta bills WhatsApp messages separately, as explained in what the WhatsApp Business API is.
Limits worth knowing
- It is not for your personal WhatsApp. It works on the company line in the official API. If you only use the app today, see the difference between the API and the app.
- Long recordings. The cap is 24 MB per file. A normal voice note is far below it; a long forwarded recording can go over and stays without text.
- Noise and accents. A very noisy recording comes out worse. If that happens often, raise the AI level for voice notes and test with real recordings.
- Language. Today the expected language is Spanish. If you serve customers in English, test before you rely on it.
- Privacy. A voice note is personal data about your customer, sometimes sensitive. Your company is accountable for how it is handled: read what data protection law asks when an agent answers on WhatsApp.
For the full walkthrough with examples of orders, complaints and team replies, see transcribing WhatsApp voice notes.
The limits
What this article does NOT answer
We would rather say it here than leave you hunting for something that is not there.
-
It does not publish an accuracy rate. We have not measured transcription accuracy with a method we can publish; test it with real voice notes from your customers before you rely on it.
-
It does not promise transcription in languages other than Spanish. The channels request the text with Spanish as the expected language, and we have not verified how a voice note in another language comes out.
-
It does not say how long voice notes and transcripts are kept. That period is not verified in this article; ask Meteor if your data policy requires it.
-
It does not cover voice notes in the history imported when you connect a number with coexistence. It only claims that new voice notes are transcribed.
-
It does not translate voice notes or summarize long calls: it transcribes conversational voice notes up to 24 MB per file.
Keep going
What this article mentions, in detail
The integrations, the Mets and the automations named above, each with its own page.
- Section
WhatsApp AI agent
The Met that reads the transcript and replies on WhatsApp, including when it hands the conversation to a person.
- Section
Pricing
The plans and the Energy included in each one, which is where the cost of every transcribed voice note comes from.
- Section
Developers
The API and webhooks where your system receives the text of each voice note without running its own transcription.
Where it comes from
Sources
Everything this article claims that we did not measure ourselves, with its origin.
- Introducing voice message transcripts WhatsApp
- Meteor API webhooks guide Meteor
- Transcribe WhatsApp voice notes on your computer Kaption
- WhatsApp voice transcription for business Transcribbit
- VozClara, AI voice transcription for WhatsApp VozClara
Frequently asked
Questions on this topic
Not directly. Meteor transcribes what comes in through a number connected to the official WhatsApp Business API. If you answer from the phone app, coexistence keeps the same number on the phone and in Meteor, when Meta enables it for your number; from then on, new voice notes are transcribed.
Voice notes that arrive through the channels are transcribed with Spanish as the expected language. If many of your customers speak another language, test it with real voice notes before you count on it.
It is deducted from your Energy balance based on length. On the Estudio to Campus plans it uses less than one US cent per minute. Every plan includes Energy, starting at US$10 on Estudio. With no balance, the voice note still arrives, just without text.
Yes. The audio stays attached to the contact and the transcript sits next to it, in the same conversation. Files that come in through the channels are kept in private storage and open through signed links, not a public address.
Yes, if the Met has a voice set up. It then replies in audio to people who wrote to it in audio, and you turn that on or off in the Met settings. If the voice fails, the customer gets the written reply.
Yes. Voice notes an agent records in the inbox, and the ones the owner sends from the phone with coexistence, are transcribed and use Energy just like the customer's. The Met's replies are not transcribed because they are already text.
The message and the audio are saved anyway and the conversation goes on. A file over 24 MB, a failure in the transcription engine or an empty balance leaves the audio without text, and the Met does not claim to understand something it could not read.
Keep reading
Other articles on the same thing
WhatsApp and channels
How to connect ChatGPT to WhatsApp Business: the real routes and Meta's rule
How do I connect ChatGPT to WhatsApp Business?
- Author
- Alejandro Kratc
- Published
- Reading time
- 6 min read
WhatsApp and channels
Unofficial WhatsApp APIs: what you risk and what Meta’s terms say
Is it safe to use an unofficial WhatsApp API to automate my business?
- Author
- Alejandro Kratc
- Published
- Reading time
- 4 min read
WhatsApp and channels
Best AI platforms for WhatsApp in Latin America (2026): seventeen compared
Which AI agent platform fits a business that sells and supports customers on WhatsApp in Latin America?
- Author
- Equipo editorial de Meteor
- Published
- Reading time
- 28 min read
Want to see it running on your operation?
We show it with your own accounts connected, not with a canned demo.