Consent before 251 MB
No model request starts until you review storage, memory, device, language, source, and license disclosures.
Audio and transcripts stay local
Turn English speech into an editable, searchable PDF on your device.
Turn a supported audio file or microphone recording into editable, timestamped English text while source audio remains on this device. After you explicitly install the speech model, you can download a searchable transcript PDF or TXT file. Recognition is English-only, does not identify speakers, and accuracy can fall with noise, overlapping voices, strong accents, or unclear recordings.
Model data
~251 MB + runtime
Explicit download, browser storage quota, and later cache eviction apply.
Device baseline
Accelerated local processing · desktop recommended
Desktop limits: 25 MB and 30 minutes.
Recognition scope
English only
Speaker labels are manual; automatic diarization is not available.
Explicit cold setup
Verified local speech model
No model request happens until you choose setup. Public model requests contain no audio or transcript. Both model file hashes must match before the model can be used.
- Model files
- Pinned and integrity-checked
- Runtime and license
- Verified local speech model · Apache-2.0 / MIT
Local input
Choose audio or record a microphone
Supported formats depend on Chromium decoding. Audio content never leaves this device.
Editable transcript
Review every timestamped segment
0 words · 0 segments
After verified local transcription, editable English text and timestamps appear here. No transcript is fabricated for silence or errors.
Integrity before inference
Downloaded speech components must match their exact SHA-256 values before activation or use.
Controllable local work
Bounded chunks expose progress, pause-at-boundary, cancellation, and retry while retaining local audio.
Searchable outputs
Edit timestamped segments and manual labels, search the transcript, then export selectable PDF text or TXT.
English recognition, not speaker identification
The local speech model is limited to English and may need cleanup for accents, names, noise, or overlapping speech. Automatic speaker diarization is not included. Speaker labels are optional text you enter yourself.
Public model assets; private source content
Explicit setup downloads the selected model and optional components. Audio, decoded samples, transcript text, edits, preview, TXT, and PDF bytes remain in this browser. Session replay is disabled on the workspace.
Continue your workflow
Related tools
These available PdfEditorOnlineFree tools are connected to audio to pdf by the tool catalog.