An interview over a soundtrack, a podcast with jingles, a vlog with music — regular transcription stumbles on such recordings. Flowmark first splits the recording into voice and instrumental and recognizes only the vocal track: the text comes out noticeably cleaner.
Promo access opens every Flowmark tool for 30 days — transcription, speech censorship, branding and more.
Recognition systems are trained on clean speech. When music plays under the voice, the text picks up gaps, invented words and whole “hallucinated” phrases — especially in pauses where only the music remains. The denser the soundtrack, the worse the result.
The classic fix is cleaning the track by hand in an audio editor before transcribing. Flowmark does it automatically: indicate that the recording contains music — and get text recognized from the clean voice.
The recording is split into stems: a vocal track and an instrumental one. Only the voice goes into recognition — the music stops polluting the text.
Diarization (who speaks), timecodes, line-by-line or paragraph output, explicit language selection — everything works in the music mode too.
The finished text arrives in the format you need, and the layout can be changed after receiving it — without re-transcribing and without extra charges.
In speech censorship the separation is always on — no mode to pick: only the vocal track is muted, and the arrangement stays intact, with no “holes” in the music.
The same bot has regular transcription, speech censorship with a custom dictionary and the rest of the media toolkit:
It depends on the density of the music. The louder and denser the soundtrack, the bigger the gain: gaps, invented words and “hallucinated” phrases on pauses go away.
Yes. Send a video with speech — the bot extracts the audio itself, separates the voice from the music and transcribes the vocal track.
Yes, the language can be set explicitly — it helps on recordings with mixed speech, strong accents or rare terminology.
A formatting button sits right next to the result: TXT to DOCX, lines to paragraphs, with or without timecodes — reformatting takes seconds and does not require repeated recognition.
Transcription is billed as usual — 1 credit per minute of the recording. There's a free plan, and the promo link opens 30 days of extended access.
Sources are stored only for the duration of processing and deleted after the result is sent; everything older than 24 hours is destroyed automatically — details in the privacy policy.
