Podcast Transcription Software: What to Buy in 2026

It depends on your workflow, and every tool on page one trades something different. Descript ($24/mo Hobbyist) wins if you also edit audio by editing text. Castmagic ($21–$29/mo entry) wins if show notes and social posts matter more than editing. Otter Pro ($16.99/mo, now capped at 1,200 min/mo) wins only for live meeting-style capture — it collapses on crosstalk panel audio. Sonix ($10/audio-hr) wins for multilingual PAYG. Rev Human ($1.99/min) is the accuracy ceiling. And if you record on Windows and want transcription with no upload and no meter, VocalFuse runs Whisper-class models locally: free tier, Pro notes $10/mo flat.

Normalize to 50 hours a year (a weekly one-hour episode). Rev AI bills $0.25/min ≈ $750/yr. Rev Human $1.99/min ≈ $5,970/yr. Sonix PAYG ≈ $500/yr. Descript Hobbyist is $288/yr billed monthly but only includes 10 media hours/mo — enough for a weekly show if transcription is all you draw. Castmagic entry tiers run $252–$348/yr with the hour library metering overage. TurboScribe is $120/yr for unlimited cloud transcription. Local VocalFuse transcription is $0/yr with no meter — the model runs on your own GPU/CPU — and Pro notes add $120/yr flat.

For podcast work, Descript — and it is not close on panel audio. Independent 2026 testing found Otter drops below 88% accuracy on four-speaker crosstalk with speaker-label confusion, while staying solid on clean two-person interviews. Descript holds ~95–96% and lets you edit the audio by editing the transcript. Otter is built for meetings: its free tier (300 min/mo, 30-min/file cap) fits one weekly episode, but the Pro cap is now 1,200 min/mo, podcast-specific features like chapter markers are missing, and SRT export is limited. Pick Otter only if you already run it for meetings and want one less tool.

If two or more people ever speak on your show, yes. Without diarization a transcript is an unbroken block of text with no names attached — unusable for show notes, quotes, or accessibility. Check that diarization is included in the base price: Sonix and Castmagic include it, AssemblyAI charges extra ($0.02/hr for speaker ID), and browser-local tools vary. Human-grade speaker attribution still beats AI on heavy crosstalk, which is one of the few remaining reasons to pay per-minute human rates on interview shows.

Free tiers worth stacking: Descript gives 1 hour/mo, Otter gives 300 min/mo (30-min/file cap), and Spotify for Podcasters auto-generates transcripts for hosted shows — but does not let you export the files. The genuinely free unlimited route is local transcription: Whisper-class models run on your own hardware at $0/minute, no account, no upload. VocalFuse ships that on Windows with SRT/VTT/TXT export; self-hosted open-source Whisper is the same engine family if you prefer rolling it yourself.

Only local tools. Every cloud option in the comparison — Descript, Otter, Castmagic, Rev, Sonix, TurboScribe, Riverside — uploads your episode to their servers for processing, with retention policies you do not control. For unreleased episodes, NDA guests, or sensitive interviews that is a real risk: your master tape becomes a copy on someone else's infrastructure. VocalFuse processes audio entirely on your Windows machine with a local Whisper model — nothing is sent to a speech API, so there is no upload wait, no file-size cap, and no vendor policy to trust.

Expect 95–97% on clean solo or two-mic audio, dropping to 88–92% on crosstalk panels, accented guests, and phone-skin recordings. Whisper-class models — the same engine family the cloud services rent you — hit ~96% on standard benchmarks and run locally, so privacy costs you nothing in accuracy. The error budget concentrates in proper nouns, brand names, and technical jargon, which is why a 10-minute proofread pass on names and terms is the standard workflow before publishing. Human transcription (~99%+) is only worth $1.99/min when accuracy is contractual.

At minimum TXT, DOCX, and SRT. SRT is the workhorse: chapter timestamps cut from it feed show notes, YouTube captions, and player chapter markers, and word-level timestamps power clip picking. VTT matters for web players. Look for tools that export SRT without gating it to a higher plan — several cap clean subtitle export on entry tiers, which is a hidden cost when you publish transcripts on your episode page.