Built for the work after transcription
A transcript is most useful when you can check it. StemVocal pairs automatic speech recognition with the original recording, word timestamps, a text editor and practical exports. You can listen to a phrase, correct it, rename a speaker and save a document or subtitle file without uploading the recording to a second service.
One StemVocal account
Our audio studio, video studio and music separator use the same credit wallet and daily free allowance. Sign in with the same email on each site. Browser sessions are separate for each domain; signing in on a new domain does not create a new wallet.
Models and limitations
Paid cloud transcription uses Whisper large-v3-turbo through faster-whisper on our GPU processor. Free cloud transcription uses the same Whisper large-v3-turbo model on Cloudflare Workers AI, and our processor when Workers AI cannot finish a file. When a free job runs in your browser instead, it uses MIT-licensed Whisper base with quantized weights in a browser worker; the model is cached after first use and audio is processed in overlapping windows, with memory independent of recording length. Optional cloud speaker labels use pyannote's speaker-diarization-community-1 pipeline, unmodified, under CC BY 4.0. Translation of paid transcripts uses Google's MADLAD-400 3B machine translation model under the Apache 2.0 license. These tools estimate words and speaker changes. They do not verify the facts in a recording or identify people. Check any important transcript against the audio.
The example recording joins two LibriVox readers from LibriSpeech (excerpts 1272-128104-0000 to 0003, read by John Rose, and 2035-147960-0000 to 0005, read by Sharon Bautista), prepared by Vassil Panayotov, Daniel Povey and collaborators. LibriSpeech is available under CC BY 4.0. We joined the excerpts with short pauses, converted the audio to MP3, added a simple visual for the video example, and show this site's own transcript of it without edits.
Software in your browser
Your browser reads recordings, and writes videos with burned-in captions, with Mediabunny 1.60.0, Copyright © Vanilagy and contributors, under the Mozilla Public License 2.0. Its complete published source package, unmodified browser module and version and file hashes are available from this site.
For a free cloud job, your browser encodes the compact audio copy with the LAME MP3 encoder 3.100, compiled to WebAssembly. LAME is free software under the GNU Library General Public License, version 2 or later (LAME's notes). It is loaded as a separate, unmodified module file that you can replace; StemVocal code only calls it. The WebAssembly build and its small C bridge come from Mediabunny's MP3 encoder package 1.60.0 by Vanilagy and contributors, under the Mozilla Public License 2.0. Its Emscripten runtime code is under the MIT or University of Illinois/NCSA license. We changed only the module's export line so browsers can load it. The LAME 3.100 source release, the encoder package with the bridge source and the version and file hashes are served from this site.
Contact
For help with a file, credit balance or purchase, email help@acapellaextractor.com. Describe the problem without attaching sensitive recordings. We will explain what information is needed to investigate.