We transcribe the speech in your browser and measure when sentences start from the audio itself; anything that can't be measured is flagged. Sentence by sentence or every few words, you pick the pace.
Your audio never leaves your computer. Only a level summary, the transcribed words and, if you pasted it, your script reach the server, and nothing is stored. The model files download once from Hugging Face. Privacy details
60
minutes of audio, max
0
bytes of your audio file uploaded
1-50
words per subtitle, or per sentence
SRT
plus VTT and a timestamp list
Workspace
MP3, WAV, M4A, FLAC or OGG. Up to 60 minutes. If you trimmed the silences in Silence Remover, use the clean file.
Paste the text of your voiceover and every word in the subtitles will be exactly yours; the transcription only finds the timing. Without it, the text comes from the audio.
Model
The model downloads once and stays in your browser. With a script, Fast is enough; without one, Balanced gets the words more accurate and Most accurate makes the fewest mistakes but takes longer.
Changing the pace applies instantly, nothing is transcribed again.
Under the hood
Speech recognition models give approximate word timings, and can place the first word after a long pause more than a second early. That's why SRT Maker measures timing from the audio itself.
Whisper transcribes your audio on your computer, word by word with timings. The audio is never uploaded.
Only a level summary and the words reach the server. Pauses are matched, in order, to the words that follow them, and those words start exactly when the speech does.
Sentence by sentence or every few words. SRT, VTT, or the timestamp list Auto Flow and Sync for CapCut read.
Next step