Transcribe WhatsApp Audio: Voice Messages to Text (2026)

To transcribe WhatsApp audio, export the chat with Including Media, upload the original ZIP to the WhatsApp audio-to-text converter, and select a voice-enabled tier. ChatToPDF matches supported audio files to their message rows, transcribes them and places each successful result back into conversation order with the exported sender and timestamp.

I built this workflow for the harder case: many voice notes scattered through one exported conversation, where sender and timing matter as much as the recognized words.

What "transcribe WhatsApp audio" actually means (and why it's harder than it sounds)

The phrase has three common intents:

1. Transcribe a WhatsApp call. A voice or video call recording is not included in Export Chat, so this guide and ChatToPDF's ZIP workflow cannot transcribe it.
2. Transcribe one saved WhatsApp audio file. A compatible single-file service can return standalone text, but it does not know the chat sender, message time or surrounding conversation.
3. Transcribe the supported voice notes included in one exported chat. This is ChatToPDF's scope: use the exported chat log as the chronology, match audio attachments to their message rows, and return successful transcripts to those rows.

The structural difference matters. A generic service sees an audio file. ChatToPDF sees an audio file plus the exported facts that identify where it belongs. The resulting PDF can therefore show the recognized speech beside the sender, original timestamp and adjacent messages instead of producing a folder of disconnected transcripts.

This guide explains that third intent. If you already understand the workflow and only need to check a ZIP and price the job, go to the batch WhatsApp audio-to-text converter.

How voice notes appear in a WhatsApp export

WhatsApp's Export Chat flow can create a message log plus available media attachments when you choose Including Media. The exact filenames and containers vary by platform and export, but push-to-talk voice notes commonly appear as Opus audio, often with .opus or .ogg naming; other supported audio forms can also appear.

The text log contains the message chronology and an attachment reference where each media message occurred. The audio file contains the speech. A useful whole-chat transcript needs both:

• the chat log to retain sender, date, time and conversation order;
• the matching audio attachment to recognize the spoken words.

Choosing Without Media gives a text-only export, so ChatToPDF may see an attachment placeholder but has no audio bytes to transcribe. Choosing Including Media asks WhatsApp to package available attachments, but it does not guarantee that every historical file will be present. Deleted, unavailable or export-limited media can still be absent. That is why the preview count is a quality-control step, not merely a marketing screen.

The codec is rarely the only accuracy variable. A compressed but clear close-mic note can be easier to recognize than a higher-bitrate recording with wind, traffic, clipping or overlapping speakers.

DefinitionOpus codec

An audio codec designed for speech and interactive internet audio at a wide range of bitrates. WhatsApp commonly uses Opus for voice notes. File extension and container can vary, so a reliable workflow checks the actual media rather than assuming every attachment is an MP3 or every .ogg file is identical.

The sender label comes from the exported message row, not voice biometrics. If another person or a television is audible inside the same note, the speech provider may recognize more than one voice, but the output is still attached to the WhatsApp sender of that message. ChatToPDF does not currently diarize multiple speakers within one voice-note file.

How the transcription router picks an engine

ChatToPDF does not depend on a single transcription engine. The current production path is a language-aware router between Deepgram Nova-3 and ElevenLabs Scribe v2. Both accept WhatsApp's Opus audio directly and return timed transcript data, but they do not have identical language coverage.

For a language Deepgram supports, the router can use Nova-3 as the primary engine. For a registry language outside Deepgram's supported set, or when the language is unknown, it uses Scribe v2. If a non-trivial note returns empty from Deepgram or that provider is unavailable, the router can retry through Scribe v2 instead of silently dropping the note. Model choice does not remove the need to review names, numbers, and important passages against the audio.

DefinitionWord error rate (WER)

The standard metric for measuring transcription accuracy: the proportion of substitutions, insertions, and deletions in the output compared with a human reference transcript. Lower is better, but a provider benchmark does not predict a specific WhatsApp note. Language, accent or dialect, microphone quality, clipping, code-switching, overlapping speakers, and background noise all change the result.

The same Nova-3 and Scribe v2 router is used for both the $49 Premium+Voice per chat conversion and the $99 Power User per chat conversion. The difference between those tiers is the total audio allowance, output bundle, and queue priority, not a promise of perfect accuracy. Clean, single-speaker recordings usually transcribe more reliably than clipped, noisy, or code-switched notes.

The $29 Premium per chat conversion does not include transcription at all — it preserves voice notes as placeholder references in the PDF without running the audio through either provider.

The pipeline works like this: your ZIP lands on ChatToPDF's server, supported audio files are extracted, and each note is sent over an authenticated HTTPS call to the provider selected by the router. Successful transcripts are then stitched back into the conversation at the corresponding positions before the PDF renders.

One deliberate choice in the pipeline: ChatToPDF does not pre-process or re-encode supported .opus audio before transcription. Nova-3 and Scribe v2 both accept the format natively, so the source audio can go directly to the selected inference endpoint.

Language coverage and accuracy across 93 languages

ChatToPDF's current registry covers 93 languages, with 57 in the excellent or high provider-published accuracy bands. Deepgram Nova-3 handles the languages it explicitly supports; ElevenLabs Scribe v2 covers the full registry and supplies the path for languages outside Nova-3's set. When no reliable language is known, Scribe v2 can identify it from the audio. You can browse the current list and planning bands in the 93-language transcription directory.

Coverage is not an accuracy guarantee. Results vary by the recording, language, accent or dialect, speaking pace, background noise, overlapping speakers, clipping, code-switching, and unusual names or numbers. Provider accuracy bands are useful for planning, but they do not predict one specific voice note. For legal, medical, financial, HR, or other high-stakes use, compare important passages with the original audio before relying on the text.

Language routing happens per note. That means one exported chat can contain notes in different supported languages without forcing the whole chat through one model. Mixed-language speech inside a single note remains harder: the dominant language signal, pronunciation, and where the speaker switches can all affect the transcript.

Worked example: Spanish voice note → text

The following is an illustrative output example, not an accuracy benchmark or a customer record. It shows how an 18-second Spanish note could appear inside the PDF when the provider returns a transcript successfully:

🎤 [Voice note — 0:18] "Hola, ¿cómo estás? Te llamo para confirmar la cita de mañana a las tres de la tarde. Si no puedes, mándame un mensaje. ¿Vale?"

The sender name would come from the matching chat row, the time would be the original WhatsApp message time, and the transcript would sit between the surrounding exported messages. The system does not derive those facts from the audio.

Spanish homophones such as haya versus halla and tubo versus tuvo illustrate why fluent-looking text can still be wrong. Names and time expressions such as a las tres de la tarde are especially important to verify when a document will be quoted or relied on.

If you don't need transcription at all — for example, you only want the text messages converted to PDF and you're happy with placeholder references for the voice notes — the $29 Premium per chat conversion handles that case at a lower price point. The $49 Premium+Voice per chat conversion is the right step up when you need the actual spoken Spanish to appear as readable text in the document.

Worked example: Hindi (mixed Hinglish) → text

DefinitionCode-switching

The linguistic practice of alternating between two or more languages within a single utterance or conversation — for example, a Hindi speaker inserting English words, phrases, or whole clauses mid-sentence (Hinglish). Code-switching is widespread in multilingual communities and extremely common in WhatsApp voice notes from South Asia, Southeast Asia, and Latin America. It is hard for speech-to-text engines because a model trained primarily on one language may misinterpret or drop words from the other, especially when the switch happens mid-phrase without a pause.

Hinglish — Hindi with embedded English words, phrases, and sometimes full clauses — is a common code-switching pattern. It is also a difficult transcription case because language detection and recognition have to follow pronunciation and vocabulary that can change mid-sentence.

This illustrative mixed-language line shows the output structure; it is not a measured accuracy result:

🎤 [Voice note — 0:22] "Yaar, kal meeting hai 3 baje, please attend karna. Project deadline aa rahi hai aur boss bahut strict hai."

An English gloss would be: "Mate, there's a meeting tomorrow at 3, please attend. The project deadline is coming up and the boss is very strict." ChatToPDF itself does not create that translation; it returns transcription in the detected source language.

Terms such as meeting, attend, project deadline and strict switch into English mid-sentence. A model can drop, respell or reinterpret those switches. If a workplace review depends on the word deadline, compare it with the audio rather than treating the generated line as independent evidence.

Sender attribution works the same as with Spanish: the name from _chat.txt appears in the PDF with the transcript, and the timestamp from the WhatsApp metadata anchors it to the correct position in the conversation.

The $49 Premium+Voice per chat conversion is the entry point for supported Hindi voice notes you want transcribed; the $99 Power User per chat conversion uses the same router with no audio-duration cap and priority processing. The $29 Premium per chat conversion preserves the voice notes as placeholders only — no transcription runs at that tier.

The $49 Premium+Voice tier — what's in it and what's not

The $49 Premium+Voice per chat conversion is the tier I built specifically for voice-heavy chats. Here is exactly what it includes and what it doesn't.

What's in the $49 Premium+Voice per chat conversion:

• Language-aware transcription — each supported note is routed between Deepgram Nova-3 and ElevenLabs Scribe v2
• A 93-language registry — 57 languages are in excellent or high provider-published accuracy bands; Scribe v2 covers the full set while Nova-3 handles languages in its supported subset
• Up to 8 hours of audio in a single chat — covers the vast majority of voice-heavy conversations; if your chat exceeds 8 hours of total recorded audio, the $99 Power User per chat conversion lifts that cap
• No message ceiling — no upper bound on the number of messages in the chat you're converting
• Sender attribution on transcripts — every transcript in the PDF carries the WhatsApp sender name from the export metadata
• Timestamps preserved — the original WhatsApp timestamp appears alongside each transcript, not the transcription time
• Three output formats — PDF, XLSX, and CSV all included; the XLSX is useful if you want to filter or sort by sender and timestamp
• Explicit cloud-processing disclosure — audio leaves the device for ChatToPDF and the provider selected for that note; Deepgram requests use its model-improvement opt-out

What's not in the $49 Premium+Voice per chat conversion:

• Real-time transcription — this tier processes already-recorded voice notes from an export ZIP; it does not monitor incoming WhatsApp messages
• Custom vocabulary lists — you cannot upload a glossary of names or technical terms; rare proper nouns can still be misheard
• Speaker identification beyond WhatsApp metadata — within a single voice note where the sender records while another person is talking in the background, both are transcribed but only attributed to the WhatsApp sender. ChatToPDF does not run speaker diarisation on the audio itself.
• Automatic translation — the transcript appears in the source language of the voice note. If a voice note is in Spanish, the transcript is in Spanish. ChatToPDF does not translate transcripts.

The tier above this one — $99 Power User per chat conversion — includes everything in $49 Premium+Voice per chat conversion plus no audio-duration cap, priority processing, and the full output bundle for that chat. It does not combine multiple chat ZIPs into one conversion.

For reference, the full tier stack: $7 Basic per chat conversion (text only, 5,000-message cap), $14 Standard per chat conversion (supported images, 25,000-message cap), $29 Premium per chat conversion (no product message-count cap, voice notes preserved as placeholders), $49 Premium+Voice per chat conversion (the 93-language router, 8-hour audio cap), and $99 Power User per chat conversion (the same router, no audio-duration cap, priority processing, and a full output bundle).

What this workflow does — and does not do

ChatToPDF is a retrospective export workflow. It processes one WhatsApp chat ZIP that the user deliberately uploads. It does not connect to a WhatsApp account, monitor incoming messages or transcribe a conversation while it is happening.

This boundary is useful when choosing a tool. If you need to read one new message in the app, try WhatsApp's native Voice message transcripts feature. If you need a live meeting or call transcribed, use a product designed for live audio. If you need a searchable record of voice notes already present in one exported conversation, use this batch workflow.

Privacy: where your audio goes and where it doesn't

This is the part I want to be specific about because the nature of voice notes — audio recordings of real conversations — means the privacy stakes are higher than with text messages alone.

Here is the exact data path for a voice note submitted through the $49 Premium+Voice per chat conversion:

Step 1 — Upload. Your ZIP file is transmitted from your browser to ChatToPDF's server over HTTPS, so the connection is encrypted in transit. The ZIP lands in a server-side job directory while extraction runs.

Step 2 — Extraction. Supported audio files are extracted from the ZIP. Each file is matched to its chat-log attachment reference by filename pattern. At this point, the extracted copy is in ChatToPDF's server-side job directory.

Step 3 — Transcription-provider call. Each supported audio file is submitted over authenticated HTTPS to the provider selected for that note: Deepgram Nova-3 or ElevenLabs Scribe v2. Deepgram requests include mip_opt_out=true, and the ElevenLabs path disables provider logging. The provider receives the audio needed for transcription, not the full chat log, and returns transcript data.

Step 4 — Storage and retention. The transcript is bundled into the output and stored in the job's server-side directory with the source export. Deletion is scheduled when files reach seven days for paid and unpaid jobs. A job still processing at that point can receive a bounded grace period of up to 24 hours so cleanup does not remove a file while a worker is reading it. If any server processing is unsuitable for the conversation, do not use the cloud workflow.

Step 5 — Delivery. The completed download is tied to the processing job and the delivery flow. Treat the job link and delivery email as sensitive: anyone with access to them may be able to reach the output.

Step 6 — Scheduled deletion. Cleanup becomes eligible seven days after the job is created. Jobs that are still processing can remain only for the bounded 24-hour grace described above; the source ZIP, extracted files, and generated downloads then become eligible for deletion.

ChatToPDF does not train a speech model on the upload. The speech provider selected for a note necessarily receives that note's audio to transcribe it. Operational systems and authorized support or infrastructure access remain part of any cloud-service risk review; do not interpret automated processing as a promise that no human could ever access server-side data.

The external-provider step matters in any privacy review. ChatToPDF controls its own pipeline, but the providers control their internal processing under their respective terms. If your voice notes contain privileged, regulated, or highly sensitive information, have the responsible privacy or legal team review both Deepgram's and ElevenLabs' current processing terms before uploading.

Edge cases: background noise, multiple speakers, voice-changing effects

Real WhatsApp voice notes are not recorded in sound-proofed studios. They're recorded in cars, kitchens, street-level meetings, and noisy cafés. Here is how those conditions can affect a transcript.

Background noise by environment.

A quiet, close-mic, single-speaker note will generally be easier to transcribe than one recorded in a busy office, restaurant, or market. Outdoors, wind and traffic can mask consonants. In a moving vehicle, road and engine noise can obscure whole phrases. Neither provider can reconstruct speech that the microphone did not capture clearly, so the audio quality going in remains the biggest practical variable.

Multiple speakers within a single voice note.

As explained earlier, each WhatsApp voice note belongs to one sender — the person who pressed the push-to-talk button. ChatToPDF attributes the transcript to that sender. If another person or a television is audible in the background, the selected provider may also transcribe that speech. The output can then interleave multiple voices while still being attributed to the WhatsApp sender.

ChatToPDF cannot currently isolate the primary speaker and discard background voices within a single audio clip. Speaker diarisation — identifying which audio segments came from which person in the same file — is not part of the current product.

Voice-changing effects.

Some WhatsApp users send voice notes with audio effects applied — pitch shifting, heavy reverb, or a deep-voice filter. Speech-recognition models are designed around natural speech, so heavily modified audio can become unreliable or fail to produce useful text.

Treat a transcript as a searchable aid, not as independent verification of the recording. If a phrase looks wrong, compare it with the source audio; do not assume fluent-looking output is exact.

WhatsApp's built-in transcripts — and fixing "Transcript not available"

Before uploading a chat, check whether WhatsApp's native feature already solves the job. WhatsApp documents this path: Settings → Chats → Voice message transcripts, choose the transcript language, then long-press one voice message and tap Transcribe. WhatsApp says generation happens on the device. That is the right route for reading one supported message inside the app.

The native and batch workflows are not substitutes. WhatsApp's published action is message-by-message and does not provide a command to create one sender-attributed PDF from all voice notes. ChatToPDF's export workflow is designed for that document job, but it requires cloud processing and a media-inclusive ZIP.

If the native action is missing or returns an error, use the dedicated WhatsApp Transcript not available guide. For the native feature and its limits, see WhatsApp speech to text. For package details and the whole-chat preview, use the WhatsApp audio-to-text converter.

FAQ

For the full chat-to-PDF workflow — including how to export on iPhone and Android, what the ZIP contains, and how all five tiers compare for non-voice conversions — see the WhatsApp to PDF guide. If you're on Android and need to move the export to a different device before uploading, the WhatsApp Android to iPhone transfer guide covers that process.

More on WhatsApp voice notes

• WhatsApp voice to text — the step-by-step workflow for turning a whole chat's voice notes into searchable text.
• WhatsApp speech to text — how WhatsApp's built-in transcription works, its limits, and when you need a dedicated tool.
• WhatsApp audio format — what the .opus codec is, how to convert voice notes to MP3, and why the format matters for transcription.
• Best WhatsApp transcription tools — an honest, category-by-category comparison of your options.