📝 Audio to Text

Fast mode produces transcript segments with segment-level timing and is the practical first pass for interviews, lectures, voice notes, and searchable reference text. Precise subtitles uses a separate alignment workflow to add word-level timing when supported. It requires another model download and more processing, so use it when synchronized highlighting or tighter subtitle cues matter rather than enabling it for every rough transcript.

Both timing modes are machine estimates. Music, overlapping speakers, background noise, accents, specialized names, and compressed recordings can shift words or boundaries. Play the source in the built-in waveform view, click timed text to spot-check alignment, and correct names, numbers, punctuation, speaker changes, and cue breaks before publishing. SRT and VTT export is a delivery format, not proof that the captions meet a broadcaster or accessibility specification.

The file picker accepts common audio and video containers supported by the browser, including WAV, MP3, OGG, WebM, FLAC, M4A, and MP4. A remote audio URL can work only when the source permits cross-origin browser access; if it does not, download a file you are authorized to use and select it locally. Recording support also depends on microphone permission and the formats exposed by the current browser.

Whisper Tiny, Base, and Small trade download size, memory, speed, and recognition accuracy. Start with a short representative clip instead of loading the largest model by default. The first run downloads model assets and later visits may reuse the browser cache until site data is cleared. Transcription inference operates in the browser and the selected media is not uploaded to OfflineTTS for recognition, while ordinary page analytics and model-host requests remain covered by the Privacy Policy.

Audio to Text does not identify speakers, translate a recording into a guaranteed target-language transcript, remove confidential information, or verify factual statements in the recording. For sensitive material, confirm that local browser processing meets your policy, use an appropriate managed device, and delete browser site data and exported files according to your retention rules. You must also have permission to record, transcribe, and distribute the source media in the relevant jurisdiction.

The selected media is processed in the browser-first transcription workflow rather than sent to an OfflineTTS upload dashboard. Model assets, ordinary page analytics, and a remote audio source that you choose may still involve their own network requests. Keep a local copy of the source and delete exported files separately if your retention policy requires it; clearing browser data does not remove files saved to your device.

TXT, SRT, and VTT are delivery formats, not evidence that a transcript is complete or accurate. Before sharing a result, compare important passages with the source, correct names and figures, and make sure you have permission to transcribe and distribute the material. For sensitive or regulated work, decide whether the browser, device, and local-file workflow meet your own policy before beginning.