Transcribing a podcast to text means converting spoken audio into a written, searchable document, and as of 2026 there are three practical routes: automated AI transcription services, human transcription services, or running open-source speech recognition models like OpenAI's Whisper on your own machine. The fastest route for most podcasters is an AI transcription tool where you upload your audio file or paste a link to the episode and receive a transcript within minutes. A one-hour episode that would take a human typist four to six hours to transcribe manually can be processed by modern AI models in two to five minutes with word accuracy rates typically between 90 and 98 percent for clear English audio.
The Direct Answer: Three Ways to Transcribe a Podcast
Also worth reading: How Accurate Is MAI-Transcribe-2 Streaming in Real-World Benchmarks? · MAI-Transcribe-2 vs Grok accuracy: Which Speech-to-Text Model Wins in 2026? · Which Free Audio to Text Tools Transcribe Most Accurately in 2026?
The first option is an AI-powered transcription service. You upload an MP3, WAV, or M4A file — or simply paste the URL of the episode from platforms like Spotify, Apple Podcasts, or YouTube — and the service returns a timestamped transcript. Services in this category include Rev's AI tier, Castos' built-in transcription, SoundWise's free unlimited tool, and browser-based tools that support 100 or more languages. Turnaround is usually under ten minutes for an hour of audio, and pricing ranges from free tiers to roughly $0.10 to $0.25 per minute of audio.
The second option is human transcription through services like Rev's professional tier or similar agencies. Human transcribers deliver 99 percent or better accuracy and handle heavy accents, crosstalk, and poor recording quality far better than machines, but they cost $1.00 to $2.00 per audio minute and take 12 hours to several days. For most podcasters publishing weekly episodes, this cost adds up quickly: a 60-minute weekly show would run $260 to $520 per month at human rates versus $15 to $60 at AI rates.
The third option is self-hosted transcription using open-source models. OpenAI released Whisper in 2022 and famously used it to transcribe more than one million hours of YouTube video during training, which tells you something about its robustness on real-world audio. Running Whisper locally on a Mac or PC costs nothing beyond compute time, and writers like those at Six Colors have documented fully automated pipelines that watch a folder for new podcast files and produce transcripts without any manual steps. The tradeoff is setup complexity: you need some comfort with the command line, a reasonably capable machine (an Apple Silicon Mac handles it well), and patience with speaker diarization, which Whisper does not do natively.
Why Podcasters Bother With Transcripts at All
There are concrete reasons transcripts have become standard practice rather than a nice-to-have. Search engines cannot index audio directly; they index text. Publishing a full transcript alongside each episode gives Google thousands of indexable words per episode, which is how shows get discovered for queries their hosts never explicitly targeted. Podcasters who publish transcripts consistently report meaningful growth in organic search traffic over six to twelve months, because every episode becomes a landing page.
Accessibility is the second driver. An estimated 5 percent of the world's population — roughly 430 million people — lives with disabling hearing loss, and deaf and hard-of-hearing listeners rely entirely on text versions of audio content. Apple recognized this when it introduced automatic transcripts for Apple Podcasts in 2024, making transcripts a native feature of the platform rather than something creators had to bolt on themselves. If Apple ships it by default, listeners increasingly expect it everywhere.
The third reason is repurposing. A transcript is raw material for show notes, blog posts, LinkedIn articles, newsletter issues, audiograms with captions, and social clips. Tools like Octopoddy, an iOS podcast app demonstrated on Hacker News, go further and use transcripts with LLMs to let listeners skip ads automatically and chat with episode content — evidence that transcripts are becoming infrastructure, not just documentation. If you ever want to build a searchable archive of your back catalog or a GPT-style assistant trained on your episodes, you need clean transcripts first.
Step-by-Step: Transcribing With an AI Service
Start by exporting your final edited audio. Most services accept MP3, WAV, M4A, FLAC, and often direct video files like MP4, so whatever your editing software (Descript, Audacity, Adobe Audition, Hindenburg) exports will work. Aim for the highest quality export you have; while modern models tolerate compressed audio, a 128 kbps or better file measurably improves accuracy on quiet passages and overlapping speech.
Next, choose your input method. Upload the file directly, or paste the episode URL if your service supports link-based ingestion — several 2026-era tools advertise exactly this workflow: paste a link, get a transcript. Link-based ingestion saves a download-and-reupload step if your episode already lives on a host or YouTube. Before committing to a paid plan, test the service against your worst-sounding episode, not your best. Accuracy varies enormously based on accent density, crosstalk, music beds, and remote interview codecs; a service that nails a solo studio monologue may stumble on a four-person Zoom panel.
Once the transcript comes back, spend ten minutes editing. Even a 96 percent accurate transcript contains roughly 25 errors per 600 words, concentrated in proper nouns, brand names, technical jargon, and homophones. Fix speaker labels, correct names, and remove filler words if your style guide calls for it. Then export in the format you need: plain text for show notes, SRT or VTT for caption files, Markdown for your CMS, or DOCX for editors. Many services now offer Markdown export directly, matching the trend seen in productivity apps like Apple Notes adding Markdown output options.
Step-by-Step: Transcribing Locally With Whisper
If you prefer not to send audio to third-party servers — relevant for shows covering sensitive topics or bound by client confidentiality — local transcription is viable. On a Mac, install Whisper via Python (pip install openai-whisper) or use a wrapper app with a graphical interface. The process is: convert your audio to a supported format (16 kHz WAV is safest), run the model with your chosen size setting, and collect the output in TXT, SRT, VTT, or JSON.
Model choice matters. Whisper comes in sizes from tiny to large-v3; the tiny model runs in near-real-time even on modest hardware but sacrifices accuracy, while large-v3 delivers the best results but needs a GPU or Apple Silicon to finish a one-hour episode in a few minutes rather than half an hour. Expect roughly 92 to 95 percent word accuracy on typical podcast audio with the large model, dropping noticeably on accented speech and multi-speaker segments. Notably, MacStories reported in 2026 that Apple's new Speech APIs now outpace Whisper for speed on Apple hardware, so macOS users should evaluate both before building a permanent pipeline.
For full automation, script the pipeline: a folder watcher detects new exported episode files, runs transcription, applies a cleanup pass (removing filler words via regex), and drops the finished transcript into your drafts folder. Six Colors documented exactly this kind of Mac automation, and once configured it removes transcription from your weekly workload entirely. The main gap remains speaker diarization — labeling who said what — which requires pairing Whisper with a separate diarization tool like pyannote.audio.
Comparing Your Options Side by Side
| Feature | AI Transcription | Human Transcription | Local Whisper |
|---|---|---|---|
| Typical accuracy (clear audio) | 90–98% | 99%+ | 92–95% |
| Cost per audio hour | Free–$15 | $60–$120 | Free (your electricity) |
| Turnaround for 1-hour episode | 2–10 minutes | 12 hours–3 days | 3–30 minutes |
| Speaker diarization | Often included | Included | Requires extra tooling |
| Language coverage | 50–100+ languages | Limited by staff | ~100 languages |
| Privacy | Audio leaves your machine | Audio leaves your machine | Fully offline |
| Handles crosstalk/accents | Weak to moderate | Strong | Moderate |
| Setup effort | None | None | Moderate to high |
Common Mistakes That Ruin Transcript Quality
The most frequent mistake is transcribing from a low-quality source. Podcasters sometimes grab a compressed stream rip or a Zoom recording at its default settings instead of the mastered export. Every additional compression generation strips consonant detail, and consonants are what speech recognition depends on. Always transcribe from your highest-quality master file.
The second mistake is skipping the edit pass entirely and publishing raw machine output. A transcript riddled with wrong names and garbled jargon reads as unprofessional and can embarrass guests who were misquoted. Budget ten to fifteen minutes per hour-long episode for corrections, focusing on names, titles, numbers, and any claims that could be legally sensitive if misrendered.
Third, many podcasters ignore timestamps and speaker labels, then wonder why the transcript performs poorly. Timestamps enable chapter markers, clickable show notes, and video captions; speaker labels make interviews readable. Both are one-click options in most services — turn them on before processing, because retrofitting them manually is miserable.
Finally, people underestimate multilingual and code-switching audio. If your show mixes English and Spanish, or features heavily accented guests, test specifically for that. Modern services advertising 100-language support handle single-language audio well, but genuine mid-sentence language switching still trips up most systems, including Whisper, which occasionally hallucinates or collapses into the dominant language.
Costs, Timing, and When to Make the Move
Pricing in 2026 spans a wide range. Free options exist: SoundWise launched a free-forever AI transcription tool for unlimited speech-to-text conversion, Apple Podcasts generates transcripts automatically for published shows, and local Whisper costs nothing. Paid AI services generally charge $0.10 to $0.25 per minute ($6 to $15 per audio hour) or offer monthly subscriptions in the $10 to $30 range that cover a typical weekly show. Human services remain at $1.00 to $2.00 per minute. If you publish one hour weekly, expect to pay nothing (self-hosted), roughly $10 to $30 monthly (AI subscription), or $260 to $520 monthly (human).
Timing-wise, the right moment to start is now, with your next episode, rather than waiting to back-catalog everything. Establish the workflow on new episodes first, then batch-transcribe your archive during a slow week — older episodes keep generating search traffic years after release, so a 100-episode back catalog is a substantial untapped asset. Note that if your show is on Apple Podcasts, transcripts may already exist for recent episodes thanks to Apple's native feature, which you can repurpose.
One caution: don't treat transcription as a magic traffic switch. A raw transcript dumped on a page with no formatting ranks poorly. Break it into sections with headings, pull out quotable moments, add a summary at the top, and structure it like an article. The transcript is the foundation; editorial effort is what turns it into a page worth ranking.
Choosing the Right Workflow for Your Show
Match the method to your show's profile. Solo or two-person interview shows with decent microphones and mostly English speech: AI transcription is the obvious default, with a light edit pass. Shows with heavy accents, field recordings, or multiple remote participants: consider AI plus human review, or budget extra editing time. Shows handling confidential material: local Whisper keeps audio on your machine. Zero-budget hobbyists: start with free tiers or Whisper on a weekend afternoon; upgrade only when time becomes more valuable than money.
Whichever path you pick, the mechanics are simple enough that your first transcript can exist within the hour. Upload or point a tool at your latest episode, review the output, fix the names, and publish it alongside your audio. Repeat weekly, automate when the manual steps annoy you, and within a year you'll have a searchable, accessible, repurposable text archive of everything your show has said.