The four-minute voice note is the nightmare of anyone who handles customers. You have to stop, find headphones and listen to the whole thing to discover the question was in the last ten seconds. Transcribing WhatsApp voice messages turns that into thirty seconds of reading, and makes the content searchable too.
There are two routes: the feature built into WhatsApp itself, and transcription outside it, which makes sense when you need volume or a searchable history.
WhatsApp's built-in voice message transcripts
Since late 2024 WhatsApp transcribes voice messages on your own device (WhatsApp blog). The feature is off by default.
How to turn it on:
- Open Settings and go to Chats.
- Turn on Voice message transcripts.
- Choose the language. The app downloads a language pack, and may wait for Wi-Fi to do it.
How to use it: long-press a voice message and tap Transcribe. The text appears under the audio. Transcription happens on your phone, and the sender does not see that you transcribed it.
Where it stops:
- It is one message at a time, by hand.
- One language at a time. A voice note in a different language from the one you picked comes out garbled.
- The text stays on that device. WhatsApp search does not find what was said inside a voice note.
- It doesn't leave WhatsApp. To move the content into a document or a system, you copy it by hand.
For a few voice notes a day, that is enough. For someone supporting dozens of customers, it is still a lot of manual work.
Bulk transcription on your computer
When volume grows, it is worth transcribing outside WhatsApp. The most widely used engine is Whisper, OpenAI's open-source speech recognition model, which handles accents and phone-quality audio well across many languages.
Option 1: run Whisper on your own machine. Free, and the audio never leaves your computer. On a Mac with Apple Silicon, the MLX build runs fast. On a PC with a GPU, the original version or faster-whisper. It assumes some comfort with a terminal.
Option 2: use a transcription API. OpenAI, Groq, ElevenLabs and others charge per minute of audio and return text in seconds. Easier to integrate, low cost per voice note, and the audio leaves your machine.
Getting the audio out of WhatsApp: from the phone you can share a voice note to an app or to your computer. For many at once, exporting the chat with media includes the .opus files next to the text. Whisper and the APIs accept opus (sometimes after converting it with ffmpeg).
Automatic transcription, the moment the voice note arrives
The next step is doing nothing at all: the voice note arrives and the transcript is already there. That takes a tool connected to WhatsApp that downloads every voice note and sends it to a transcription engine.
Platforms built on the official API sometimes include it. With open-source connectors (Evolution API, WAHA, Baileys) you wire it yourself with an n8n flow or a short script: on incoming audio, transcribe and save the text. We compare those connectors in open-source WhatsApp APIs and WAHA vs Evolution API.
What changes once audio is text
- You read instead of listening. Response time drops, especially on long voice notes. More on that in WhatsApp response time.
- Decisions become findable. "The client approved the color in Tuesday's voice note" becomes a search, not a hunt.
- AI understands the conversation. An AI agent that reads the history needs the voice note as text, or it will answer as if the question was never asked.
Things to be careful about
- Privacy. Client audio can contain personal data. If you send it to an external API, read the provider's retention policy. Running locally avoids the question.
- Check names and numbers. Transcription gets proper names, amounts and dates wrong now and then. Before acting on a number that came from a voice note, listen to that part.
- Keep the original audio. The transcript is great for reading and searching, but the audio is still the source. When tone or a number is in doubt, the audio wins, so don't delete it after transcribing.
- Several speakers. Whisper does not separate speakers by default. For recorded meetings, look for tools with speaker identification.
FAQ
Does WhatsApp transcription work on desktop? The built-in feature launched on phones. On a computer, the workaround is downloading the voice note and transcribing it with another tool, such as Whisper.
Does the sender know I transcribed their voice note? No. According to WhatsApp, transcription happens on your device and is not shown to the sender.
Is Whisper free? The model is open source and runs free on your machine. What costs money is using it through an API, billed per minute by the provider you choose.
What about voice notes in groups? Same thing. In a client group, transcribing voice notes is what keeps decisions findable later. Once audio is text, you can also summarize the whole chat with AI.
Where WACrew fits
WACrew is an open-source agent for WhatsApp customer service that transcribes voice notes as soon as they arrive: with Whisper large-v3 through MLX on a Mac, or through an API on other systems. The transcript is stored next to the conversation, and search finds text inside voice notes, including through the MCP tools your coding agent uses. The code is on GitHub.



