Phone Call Transcription: Record, Upload, Get Text
There are three ways to get a phone call into text: record on your device and upload the file (works for any call you already have), let a VoIP platform transcribe live (requires switching phone systems), or use a dedicated recorder app. This page covers the first path end-to-end — including the two things most guides skip: whether recording is legal where you are, and the honest accuracy numbers for phone audio, which is measurably the hardest thing any transcription engine faces.
By VexaScribe Editorial · Updated
The Fastest Path: Transcribe a Call You Already Recorded
If the recording exists, you're four steps from a searchable transcript with speaker labels.
Get the file off the phone
iPhone native recordings live in the Notes app (share → save the audio). Android recordings land in the Phone or Recorder app. VoIP recordings download from the admin console. Any format works: MP3, WAV, M4A, AAC, AMR, OGG.
Upload — or paste a link
Drag the file into VexaScribe, or paste a Google Drive share link or any direct HTTPS audio URL if the recording is already in the cloud. Files over 25MB from Drive work too.
Let diarization label the speakers
A two-person call is the ideal case for speaker detection — expect clean "Speaker 1 / Speaker 2" separation, then rename them to "Client" and "You" in the editor.
Search, chat, export
Ask the AI chat "every time pricing was mentioned" or "what did they commit to?", jump to timestamps, and export TXT/DOCX for the file or SRT if the call becomes content.
Cost reference: a 30-minute call costs about $0.15 on the Starter plan ($2/mo for 200 minutes) — full math on the cost calculator.
How to Record Calls in 2026
The recording landscape changed materially in late 2024 when Apple shipped native call recording. Here's every mainstream path and its catch.
| Method | Requires | You get | The catch |
|---|---|---|---|
| iPhone native Call Recording | iOS 18.1+, supported region | Audio + transcript in Notes app | All parties hear an automatic announcement; not available in the EU, Turkey, Saudi Arabia, and ~15 other regions |
| Android built-in recorder | Pixel / Samsung Phone app, region-dependent | M4A/AMR file on device | Availability varies wildly by country and carrier; most versions announce recording to both parties |
| Google Voice | Free US Google Voice number | MP3 in your Voice account | Records incoming calls only (press 4); announces recording; US-only |
| VoIP platform recording | Zoom Phone, Teams, RingCentral, etc. | MP3/MP4 in admin console | Best audio quality of any method (16 kHz+); needs workspace admin to enable |
| Speakerphone + voice memo | Any second device | M4A voice memo | Universal fallback — works everywhere, but room echo cuts accuracy noticeably |
iPhone details verified July 14, 2026 against Apple's official guide. Apple's built-in transcript is genuinely good for supported languages — upload to VexaScribe when you need more languages (90+), speaker-labeled exports, or AI chat across many calls.
Is It Legal to Record Phone Calls?
The rules attach to recording, not transcription — transcribing a legally made recording is fine. The recording rules split into two regimes:
One-party consent
US federal law and the majority of US states require only that one participant (you) consents. If you're on the call, you may record it.
All-party consent
Roughly a dozen US states — including California, Florida, Illinois, Maryland, Massachusetts, Pennsylvania, and Washington — require every participant's consent. Cross-state calls: follow the stricter rule.
The rule that always works: announce it and get the yes on tape — "I'd like to record this call so I can focus instead of taking notes — is that OK?" That satisfies every US regime, and it's why Apple's and Google's native recorders play an automatic announcement. In the EU, GDPR treats call recordings as personal data: you need a lawful basis, informed participants, and a retention plan — announcement alone may not be enough for business calls. This is general information, not legal advice; check your jurisdiction's current rules for anything sensitive.
Why Phone Audio Is the Hardest Thing You'll Ever Transcribe
Cellular calls are compressed to 8 kHz bandwidth — half the frequency range of normal recordings — and layered with codec artifacts, crosstalk, and casual overlapping speech. The result shows up brutally in benchmarks: on CallHome (the standard conversational-telephone test set), the best engines still get roughly one word in five wrong.
| Engine | CallHome WER | In practice | Note |
|---|---|---|---|
| Deepgram Nova-3 | 21.8% | ~1 in 5 words wrong | Best measured on phone audio |
| AssemblyAI Universal-2 | 23.4% | ~1 in 4 words wrong | Strong entity handling helps names survive |
| Whisper Large-v3 | 26.4% | ~1 in 4 words wrong | The engine class VexaScribe runs |
Same benchmark data as our Deepgram and Whisper accuracy pages. Important context: CallHome is casual conversational phone audio — structured business calls on decent connections transcribe far better, typically 80–90%+.
What actually improves phone transcription accuracy
| Fix | Impact | Why it works |
|---|---|---|
| Record on VoIP instead of cellular | Largest single improvement | VoIP captures 16 kHz+ wideband audio; cellular calls are compressed to 8 kHz — half the frequency information |
| Use headsets, not speakerphone | 5–10 WER points | Speakerphone adds room echo and distance — the far-field problem that degrades every engine |
| Record both sides on separate channels | Cleaner speaker labels | Stereo call recordings let diarization assign channels instead of guessing at crosstalk |
| Skip the call-quality apps that re-compress | Avoids stacked compression | Each lossy re-encode (AMR → MP3 → upload) compounds; keep the original file format |
Three Real Workflows
Sales team: weekly call review
Reps record via the VoIP platform → upload the week's calls in bulk → AI chat asks each transcript "what objections came up?" and "what pricing was discussed?"
Honest expectation: VoIP audio transcribes at roughly 90%+ accuracy — good enough that quotes need a spot-check, not a re-listen.
Journalist: source call on deadline
Record (with stated consent, on tape) → upload the file directly or via URL → pull quotes with timestamps, then verify each against the audio before publishing.
Honest expectation: Cellular audio runs ~75–85% accurate. Every quote you print still gets checked against the recording — the transcript is for finding them, not trusting them.
Legal intake: call log for the case file
Record intake calls (check your state's consent rule first) → transcribe with speaker labels → store transcript + audio together; search across calls when the matter develops.
Honest expectation: AI transcripts work as searchable working notes. Anything filed with a court needs certified human transcription — see our legal transcription guide.
When You Need Something Else
You need live transcription during the call — that's a VoIP platform feature (Dialpad, Zoom Phone, Teams), not an upload workflow. VexaScribe transcribes recordings, not live lines. If your whole team needs in-call transcripts, switching phone systems is the honest answer.
The transcript goes to court — filings need certified human transcription (~$1.58–$1.99/min). Use AI for the searchable working copy and pay for certification only on the calls that matter. Details in our legal transcription guide.
It's a voicemail, not a call — iPhone Live Voicemail and Pixel already transcribe voicemails natively; see when uploading beats built-in on our voicemail transcription page.
Have a call recording right now?
Upload it or paste a link — speaker labels, timestamps, AI chat over the transcript, 90+ languages. From $2/mo.
Transcribe a Call FreeFrequently Asked Questions
How do I transcribe a phone call I already recorded?
Get the file off the device — iPhone native recordings save to the Notes app, Android recordings to the Phone/Recorder app, VoIP recordings download from the platform console — then upload it to VexaScribe or paste a Google Drive link or direct HTTPS URL. All common phone formats work: MP3, WAV, M4A, AAC, AMR, OGG. The AI transcribes with speaker labels and timestamps in a few minutes; a 30-minute call costs about $0.15 on the Starter plan.
Is it legal to record and transcribe phone calls?
Transcribing a legally made recording is fine — the legal question attaches to the recording. US federal law and most states require one-party consent (you, as a participant, may record). Roughly a dozen states — including California, Florida, Illinois, Maryland, Massachusetts, Pennsylvania, and Washington — require every participant's consent; for cross-state calls, follow the stricter rule. In the EU, GDPR adds lawful-basis and retention requirements for business calls. The universally safe move: announce the recording and get agreement on tape. This is general information, not legal advice.
How accurate is phone call transcription really?
Phone audio is measurably the hardest common scenario for every transcription engine. On the CallHome conversational-telephone benchmark, the best engines score 21.8–26.4% word error rate — roughly one word in four or five wrong — because cellular audio is compressed to 8 kHz bandwidth with codec artifacts and crosstalk. Structured business calls on good connections do far better, typically 80–90%+. The biggest single improvement is recording on VoIP instead of cellular; avoiding speakerphone is second.
Can I record calls natively on iPhone now?
Yes — since iOS 18.1 (late 2024), the iPhone Phone app records calls natively: tap the waveform icon during a call, all participants hear an automatic announcement, and the recording saves to the Notes app with an on-device transcript. It isn't available in all regions (excluded: the EU, Turkey, Saudi Arabia, and roughly 15 others). Apple's built-in transcript is good for supported languages; upload the audio to VexaScribe when you need broader language support (90+), speaker-labeled exports, or AI chat across many calls.
Does phone call transcription identify who is speaking?
Yes. A two-person call is close to the ideal case for speaker diarization — expect clean Speaker 1 / Speaker 2 separation, which you can rename to 'Client' and 'You' in the editor. Accuracy drops when speakers talk over each other, which is common in heated or casual calls. If your recording setup captures each side on a separate stereo channel (some VoIP platforms do this), speaker attribution gets significantly more reliable.
How do I record a call on Android?
It depends heavily on region and manufacturer. Pixel and Samsung Phone apps include call recording in some countries (with an automatic announcement to both parties), but carriers and local law disable it in many others. Alternatives: Google Voice records incoming calls free in the US (press 4 during the call), VoIP apps record natively, and the universal fallback is speakerphone plus a voice memo on a second device — it works everywhere but costs accuracy through room echo.
Can I transcribe both sides of the call, or just my voice?
Any real call recording (native recorder, VoIP, Google Voice) captures both sides in one file, and transcription covers everything spoken. The exception is the speakerphone + voice-memo fallback: the far party comes through the phone's speaker, so their audio is quieter and echoier — expect noticeably worse accuracy on their side of the conversation. If both sides matter equally, use a method that records the line directly rather than the room.
How much does phone call transcription cost?
On VexaScribe: a 15-minute call ≈ $0.08, a 30-minute call ≈ $0.15, an hour ≈ $0.30 — plans start at $2/month for 200 minutes. Human transcription of the same hour runs $95–$120 ($1.58–$1.99/min at GoTranscript/Rev rates, verified July 2026) and makes sense mainly when the transcript becomes a court filing. Full comparison math on our transcription cost calculator.