Audio to Text with Whisper AI — Free, Local & Private
Transcribe any audio file to verbatim text — entirely in your browser, with no upload and no API key required. Powered by OpenAI's Whisper running via WebAssembly, this tool processes your audio locally on your device. Output includes [MM:SS] timestamps, speaker labels, and full verbatim content. Download as TXT or DOCX — completely free.
Verbatim Transcript
How to Use Whisper AI Transcription
This tool uses OpenAI's Whisper — an open-source automatic speech recognition model — running directly inside your browser via WebAssembly. Unlike cloud transcription services, nothing is sent to any server. Your audio file stays entirely on your device from upload to download.
Step 1 — Download the Whisper Model
Before transcribing, select a model size and click Load Model. The model file is downloaded from a CDN to your browser cache — this is a one-time step. On subsequent visits the model loads instantly from cache. Choose your model based on your needs: Tiny (~39 MB) for speed on clear audio; Base (~74 MB) for a good balance; Small (~244 MB) for the highest accuracy on difficult recordings.
Step 2 — Upload Your Audio
Click Choose Audio or drag and drop your file. Supported formats include MP3, WAV, M4A, OGG, FLAC, OPUS, WEBM, AAC, and MP4. There is no file size limit — recordings longer than ~10 minutes are automatically split into overlapping 30-second chunks, transcribed in sequence, and merged into a single continuous output.
Step 3 — Configure Language and Speakers
Select the language of your audio from the language grid. Then set the number of speakers and enter their names — for example, "Interviewer" and "Respondent", or "Dr. Chen" and "Participant". These labels appear throughout the transcript to identify who is speaking at each point.
Step 4 — Choose Output Format and Transcribe
Select your preferred output format:
- Verbatim + Timestamps — Full verbatim output with [MM:SS] timestamps and speaker labels. Ideal for qualitative research coding in ATLAS.ti, NVivo, or MAXQDA.
- Clean text only — Plain transcript without timestamps, for easy reading, editing, and sharing.
- SRT subtitles — Standard subtitle format for use with video players and editing software like Premiere Pro or DaVinci Resolve.
Click Transcribe. A progress bar shows real-time processing status. When complete, the transcript appears in the editable output box — you can correct any errors directly before saving.
Step 5 — Edit and Download
The output box is fully editable. Correct any transcription errors, anonymise speaker names, or remove identifying information directly in the browser. Then click ⬇ TXT to download a plain-text file, or ⬇ DOCX to download a formatted Word document with speaker labels and timestamps preserved.
Who Is This Tool For?
- Qualitative researchers — Transcribe in-depth interviews, focus groups, and oral history sessions for coding and thematic analysis.
- Journalists — Convert recorded source interviews to text for quoting, fact-checking, and archiving.
- Students — Transcribe lectures, seminars, or group discussions for study notes.
- Podcasters — Generate show notes and episode transcripts from recordings.
- Legal and medical professionals — Transcribe sensitive recordings that must never be uploaded to a third-party server.
- Anyone needing a quick, private, no-cost way to turn audio into text.
Frequently Asked Questions
Is this tool suitable for qualitative research verbatim?
Yes. The output is designed specifically for verbatim research transcripts: each segment includes a [MM:SS] timestamp, a speaker label (e.g., "Interviewer:", "Respondent:"), and the full verbatim content. You can edit speaker names and correct transcription errors directly in the output box before downloading.
Is my audio uploaded to a server?
No — your audio never leaves your device. This tool uses Whisper AI running entirely inside your browser via WebAssembly. There is no server, no API call, and no network connection during transcription. Your recordings remain 100% private.
Do I need an API key or account?
No API key or account of any kind is required. The Whisper AI model is downloaded once to your browser cache and runs completely offline after that. Everything is free, with no sign-up, no login, and no usage limits.
What is the first-time model download?
Before transcribing, the Whisper model file must be downloaded to your browser cache. Depending on the model you choose, this is ~39 MB (Tiny), ~74 MB (Base), or ~244 MB (Small). This download happens only once — on subsequent visits the model loads instantly from your browser cache.
Which model size should I choose?
Tiny (~39 MB) is fastest and works well for clear, single-speaker audio. Base (~74 MB) offers a good balance between speed and accuracy. Small (~244 MB) delivers the best accuracy, especially for accented speech, multiple speakers, or noisy recordings — but takes longer to process. Start with Base if you are unsure.
What audio formats are supported?
MP3, WAV, M4A, OGG, FLAC, OPUS, WEBM, AAC, and MP4 are supported. For best transcription accuracy, use clear recordings at 16 kHz or higher sample rate. If your audio has background noise, consider using the Audio Compressor to normalize it first.
Is there a file size or duration limit?
There is no file size limit — the tool runs entirely in your browser. Long recordings are automatically split into overlapping 30-second segments, transcribed in sequence, and merged into one continuous transcript. Files up to several hours in length can be processed, depending on your device's memory and CPU.
How accurate is the transcription?
Accuracy is high for clear speech in supported languages. Results may be lower for heavy accents, overlapping speakers, very quiet or noisy recordings, or domain-specific jargon. The transcript is fully editable in the output box — correct any errors before downloading. Using the Small model gives the best accuracy.
What are the output formats?
Verbatim + Timestamps: [MM:SS] Speaker: text — ideal for qualitative research coding in Atlas.ti, NVivo, or MAXQDA. Clean text only: plain transcript without timestamps, for easy reading and editing. SRT: standard subtitle format compatible with video players and editing software.
Can I use this for research ethics compliance?
Yes — because audio never leaves your device, this tool is suitable for sensitive interviews and compliant with most institutional ethics requirements. No third-party service ever receives your data. Always inform participants that AI transcription will be used. The editable output lets you anonymise speaker names and remove identifying information before saving.