WhatsApp Voice Message Transcription — Free, .opus Ready

Two paths. (1) Native (built-in, limited): WhatsApp added an in-app voice message transcript feature in April 2024. Open a voice message → hold → Transcribe. Requires iOS 16+ or a recent Android version, and only works for English, Spanish, Portuguese, Russian, Hindi and Italian as of July 2026. The transcript displays inline but cannot be exported. (2) VexaScribe (any language, any device, exportable): save the .opus voice note from the WhatsApp chat, upload it to VexaScribe, get back TXT / SRT / VTT / JSON / CSV. Runs on Whisper Large-v3 (99 languages, 93-95% precision on clean audio in Tier 1 languages). 30-minute free trial, no credit card required.

iOS: open the chat, tap and hold the voice message → Share → Save to Files → pick a folder. The file saves as .opus or .ogg. Android: open the chat, tap the three-dot menu on the voice message → Share → Files (or your file manager) → save. Alternatively, the WhatsApp folder in your phone's internal storage (Android/media/com.whatsapp/WhatsApp/Media/WhatsApp Voice Notes/) contains all received voice notes as .opus files. Desktop (WhatsApp Web / Desktop app): right-click the voice message → Save audio as → save as .ogg or .opus. Fastest workaround for any platform: forward the voice message to your own phone number or a saved-messages chat, then use the file manager to grab it.

WhatsApp uses the Opus audio codec (IETF RFC 6716) at 16 kHz sample rate for voice messages because Opus delivers speech-quality audio at much lower bitrates than MP3 — around 12-16 kbps for clear speech vs 64-128 kbps for MP3 of similar quality. That saves WhatsApp's servers and users' mobile data significantly across the billions of voice notes sent daily. Downside: many older transcription tools don't support .opus and force you to convert first. VexaScribe handles .opus natively — no conversion step, no quality loss from re-encoding.

Only if you use English, Spanish, Portuguese, Russian, Hindi or Italian, as of July 2026. WhatsApp rolled out native transcripts in April 2024 with English, Spanish, Portuguese, Russian and Hindi, and added Italian later. French, German, Arabic, Japanese, Chinese, Korean, Turkish, Indonesian and 90 other languages are not natively supported yet. If your voice message is in an unsupported language, WhatsApp shows nothing — you'll need an external tool. VexaScribe uses Whisper Large-v3 which covers 99 languages including all the WhatsApp-heavy markets: Hindi, Hinglish (Hindi-English code-switching), Arabic (Egyptian, Levantine, Gulf variants), Turkish, French, German, Japanese, Mandarin, Bengali, Tamil, Punjabi.

No. WhatsApp's in-app transcript displays as a bubble beneath the voice message but has no export button, no copy-to-clipboard for the full transcript in some builds, and no format options (no SRT, no timestamps). If you need the text as a file — for archiving customer support conversations, quoting a source in an article, generating subtitles for a video repurposing the voice, or feeding the text into another tool — you have to type it out manually or use an external tool. VexaScribe exports to TXT (plain text), SRT (timestamped subtitles), VTT (HTML5-compatible subtitles), JSON (developer format with per-word timestamps), and CSV (analysis).

Yes. Whisper Large-v3 processes multi-minute audio at the same accuracy as short clips — there's no quality degradation on longer files. WhatsApp voice notes have a hard cap around 100 minutes per message on modern versions. On the free 30-minute trial, one long voice note eats into your minute quota; the Starter plan (200 min/month, $2 equivalent) covers moderate use, Basic (1000 min/month, $5) covers heavy WhatsApp-first workflows (journalism, community management, family archives).

Reasonably well. Whisper Large-v3 was trained on multilingual audio including code-switching examples, so Hindi-English (Hinglish), Portuguese-Spanish (Portuñol), French-English (franglais), and Spanglish are transcribed with a single language chosen by the model — usually the dominant language of the clip. If the message is 60% Hindi and 40% English words mixed in, expect Hindi script output with English words transliterated or written in Latin script inline. For strict per-word language tagging you'd want a specialized code-switching model (WhisperX with word-level alignment is closer). For most journalists and community managers, Whisper Large-v3's default handling is good enough to read and quote accurately.

Your file is uploaded to our EU infrastructure (AWS eu-west-2, London), transcribed by Whisper Large-v3, and returned to you. We do not retain the file beyond processing, do not use it to train models (contractual non-reuse), and do not share it. For sensitive content (customer support recordings with personal data, source interviews under confidentiality, medical or legal voice notes), the contractual non-reuse guarantee is the honest baseline — for HIPAA-covered clinical content specifically, we're not HIPAA-certified, so use a HIPAA-covered vendor instead. See our privacy policy for details.

Only if you recorded the call. WhatsApp voice and video calls are end-to-end encrypted and not recorded by WhatsApp — you cannot 'download' or transcribe a past call from WhatsApp's servers because it doesn't exist there. What you can do: (1) On iPhone, use the built-in Voice Memos app during the call by recording on a second device; (2) On Android, use built-in call recording (Pixel, Samsung One UI 5+) or a third-party recorder — check local laws first, several jurisdictions require both-party consent. Then upload the resulting audio file (MP3, M4A, WAV, .opus) to VexaScribe. For multi-speaker call transcription with speaker labels, see our speaker labeling / diarization guide.

Whisper Large-v3 achieves 93-95% word accuracy on clean audio in Tier 1 languages (English, Spanish, French, German, Italian, Portuguese, Dutch, Japanese, Mandarin, Korean, Russian, Polish) per the FLEURS benchmark. Tier 2 languages (Arabic, Turkish, Hindi, Vietnamese, Thai, Indonesian, Ukrainian, Czech, Hungarian, Romanian) hit 88-92%. Tier 3 (Swahili, Bengali, Tamil, Welsh, lower-resource) hit 75-88%. WhatsApp voice messages recorded in a quiet room with a modern phone microphone consistently land in the Tier 1 clean-audio range. Voice notes recorded outdoors, in cars, or on speakerphone drop by 3-8 percentage points depending on noise floor.