Speech to Text — DOCX transcript

Typical accuracy is 90–98% on clear speech. Background noise, heavy accents and overlapping speakers reduce accuracy — a short cleanup usually gives you a broadcast-quality transcript.

MP3, WAV, M4A, AAC, FLAC, OGG, OPUS and most other common formats. The tool converts them automatically before transcription.

Up to 2 hours per upload in the free tier. Longer files can be split, or use the batch endpoint for archives and podcast back-catalogues.

Yes. Uploads are encrypted in transit, stored only for the duration of the job, then permanently deleted. We never train on your audio.

English, Spanish, French, German, Portuguese, Italian, Russian, Chinese, Japanese, Korean, Vietnamese, Hindi, Arabic and 80+ more.

The model handles clinical terminology, legal statutes and engineering jargon well — Word Error Rate stays under 5% on clear audio. Proper nouns, acronyms and drug/case names are preserved verbatim.

No. Punctuation, sentence breaks and paragraph structure are added automatically. Speaker labels and timestamps are inserted where relevant, so the output is ready for review or publishing.

Speaker diarisation labels each participant separately (Speaker 1, Speaker 2…). Background noise is filtered before recognition, and the engine copes well with meetings, interviews and phone calls.

Uploads and transcripts are deleted from our servers as soon as your download finishes. Nothing is kept for training, analytics or any third party.

TXT, SRT, VTT, DOCX, PDF, JSON and CSV. Every job lets you switch formats without re-uploading.