WhatsApp Business API Voice Transcription: July 2026

You've been through the WhatsApp Business API reference twice, looking for the endpoint that turns an inbound voice note into text. It isn't there. It has never been there. That's not an oversight in your reading - it's the single most important fact about WhatsApp Business API voice transcription, and almost nothing written about the topic says it plainly.

Here's what the platform actually gives you, what it quietly withholds, and the two traps that silently break voice messages for teams who don't know about them.

Last verified July 2026 against Meta's WhatsApp Business Platform documentation, including the changes shipped in March and June 2026.

The Short Answer

There is no WhatsApp Business Platform endpoint that transcribes an inbound voice message. Not in the Cloud API, not in the On-Premises API. When a customer sends your business a voice note, the webhook hands you a media ID. You call the media endpoint, you download an audio file, and from that point you are entirely on your own.

Turning that audio into text is your problem to solve, with speech-to-text of your own choosing and at your own cost.

The transcripts that WhatsApp users see in their chats do exist - but they're generated on the recipient's phone and never leave it. They are not transmitted to WhatsApp, and they are certainly not transmitted to you. Even if a human agent on your team reads a transcript inside the WhatsApp Business app, that text is stranded on that one handset. There is no API that will hand it to you.

For the non-developer side of this - setup, languages, troubleshooting - see our WhatsApp Business transcription FAQ.

The Irony: Meta Will Transcribe a Call, But Not a Voice Note

This is where the platform gets genuinely strange.

On 30 June 2026, Meta shipped call transcription for the WhatsApp Business Calling API. It is not a token feature. It auto-detects the spoken language across roughly 54 languages, separates the business speaker from the customer, and returns word-level timings with per-word confidence scores as structured JSON.

Meanwhile, voice message transcripts on Android still support four languages: English, Portuguese, Spanish and Russian. Unchanged since November 2024. We've broken down which languages WhatsApp transcription actually supports if you need the full picture per platform.

So the platform can transcribe Tamil, Swahili, Vietnamese and Urdu - on a phone call. It just won't do it for the voice note your customer sent you thirty seconds earlier. Whatever the reason, the practical consequence for you is simple: the engine exists, and you cannot point it at your voice notes.

How call transcription actually works

• Opt in per call by passing a transcription object when you initiate or accept the call. It isn't on by default.
• Consent is mandatory and audible. Meta mixes a spoken announcement into both audio streams before transcription begins, and you must supply a stated purpose. You cannot transcribe a call silently.
• Delivery is asynchronous. You get a webhook with a media ID, then download the JSON transcript.
• You have seven days. After that the transcript is no longer downloadable. If you need to keep it, you must pull it and store it yourself, promptly.
• It's currently free, with Meta's docs stating that separate pricing is planned but not yet dated. Don't build a business case on the price staying zero.

Two Traps That Silently Break Voice Messages

Trap 1: it must be .ogg with the OPUS codec

Since October 2025 you've been able to send audio as a true voice message - waveform, play button, the lot - rather than as a file attachment. But there is a hard requirement buried in the docs.

Voice messages must be .ogg files encoded with the OPUS codec. Meta's documentation is unambiguous: "If you send a different file type or a file encoded with a different codec, voice message transcription will fail."

What makes this vicious is the failure mode. There's no error and no warning - the message sends, it plays, and the transcript just never appears. Teams lose days to this, combing through their webhook handling when the actual bug is one line in their encoding step.

Trap 2: the recipient decides, not you

Even with a perfectly encoded OPUS file, you do not control whether a transcript appears. The recipient does. Every WhatsApp user sets voice message transcripts to one of three states:

• Automatic - the text transcription appears with the message.
• Manual - they see a "Transcribe" option and choose whether to use it.
• Never - no text appears, ever.

There is no API parameter that overrides this, and no way to query which setting a given user has. If your product experience depends on the customer reading a transcript of your voice message, understand that you are depending on a setting you cannot see and cannot change. Design accordingly - if the content matters, send text.

What Did Land in 2026

The "played" webhook

From 17 March 2026, voice messages sent by a business receive a "played" status webhook the first time the recipient plays them. Until now, "delivered" and "read" were the end of the story, and a read receipt on a voice message told you very little.

This is a genuinely useful signal. It's the first time you can distinguish a voice message that was merely delivered from one that was actually listened to - which is the only thing you really wanted to know.

Voicemail arrives as an ordinary audio message

Voicemail reached general availability on 25 June 2026. The detail worth internalising is how it's delivered: voicemails come through the messages webhook as inbound audio messages.

Which means they land in exactly the same position as every other voice note - as a media ID pointing at an audio file, with no transcript attached. Your missed calls now generate more audio for you to transcribe yourself.

So What Do You Actually Build?

Strip away the detail and the architecture is forced. For any inbound voice note, you receive a media ID, you download the audio, and you bring your own speech-to-text. There is no supported path around this.

That leaves you with a real decision, and it's worth making deliberately rather than by default:

Build it. Wire the media endpoint to a speech-to-text provider. You'll own the pipeline, the retries, the audio storage and its retention policy, the per-minute costs, and the ongoing question of accuracy on real-world audio recorded in vans and cafes. Reasonable if transcription is core to your product and you want control over the model.

Don't build it. If you just need your team to read what customers said, standing up and maintaining a transcription pipeline is a lot of engineering to reinvent something that already exists. Forwarding voice notes to a transcription service gets you text in the chat, in over 50 languages, with no infrastructure and no audio for you to store and secure.

Be honest about which problem you have. Plenty of teams have built the pipeline because the API's silence made it feel mandatory - not because transcription was ever the thing they wanted to be good at.

Using the WhatsApp Business app rather than the API? Different product, different rules - the transcript lives on your phone and the setup is a settings toggle. See our guide to transcribing WhatsApp Business voice messages instead.

Sources

Checked against Meta's primary documentation in July 2026.

• Meta for Developers - Audio messages - the .ogg/OPUS requirement, the recipient's Automatic/Manual/Never setting, and the "played" status webhook from 17 March 2026.
• Meta for Developers - Call transcription - opt-in flow, the mandatory consent announcement, JSON transcript structure, language coverage and the seven-day download window.
• WhatsApp Help Centre - About voice message transcripts - the on-device, recipient-side behaviour and the Android language list.

Frequently Asked Questions

Is there a WhatsApp Cloud API endpoint that transcribes an inbound voice note? No. As of July 2026, no endpoint on the WhatsApp Business Platform returns a transcript of an inbound voice message. You get a media ID, you download the audio, and the transcription is yours to solve.

Can a business see the transcript of a voice message it received? Not through the API. Transcripts are generated on the recipient's device and never leave it. Even when your agent reads one in the WhatsApp Business app, that text stays on that phone.

Why does my WhatsApp voice message never show a transcript? Almost always one of two things. Either the file isn't .ogg encoded with OPUS - in which case transcription fails silently - or the recipient has set voice message transcripts to Manual or Never, which you cannot see or override.

Does the WhatsApp Business Calling API support transcription? Yes, since 30 June 2026. Opt in per call, both parties hear a required announcement, and you get JSON with speaker separation, word-level timings and confidence scores across roughly 54 auto-detected languages. The transcript is downloadable for seven days. It transcribes calls, not voice notes.

Is call transcription free? It is at the moment. Meta's documentation says separate pricing is planned, with no launch date given. Treat the current price as temporary.