Topic: Lyric transcription

[meltycrayon]

Probably like most of you, if I'm prepping a recording and can't identify a song, I'll Google snatches of lyrics in hopes of getting a hit. But sometimes, with a recording's sound quality and/or my poor hearing, I can't understand enough of the lyrics to come up with a Google-able phrase. So I've tried a lot of the online AI transcription sites, and the best by far has been https://elevenlabs.io/audio-to-text

If you've had great experiences with other sites, please speak up as maybe next time elevenlabs might let me down.

[Gutbucket]

Interesting, thx.  Good resource. Its an obvious application I suppose, although I've not tried using anything to do this yet.

[morst]

Sometimes if I can't understand the lyrics over the music, I'll do a band-pass type filter arrangement with EQ, and cut high and low frequencies, leaving mostly the human vocal range.

A lot of vocal energy is concentrated in the range from approximately 800 - 2000 cycles so I often start by cutting everything else.

I won't say it works great, but it's helped me a few times.


edit to add: I search the web at large for lyrics, but also search some keywords on https://setlist.FM and https://Discogs.Com, and browse shows on https://etreedb.org/ as well, of course, as the archive!

[aaronji]

It can also be helpful to check recent set-lists. A lot of bands have a pool of songs that they draw from for a given tour, which can help narrow down the options.

[AbbyTaper]

Sometimes if I can't understand the lyrics over the music, I'll do a band-pass type filter arrangement with EQ, and cut high and low frequencies, leaving mostly the human vocal range.

A lot of vocal energy is concentrated in the range from approximately 800 - 2000 cycles so I often start by cutting everything else.

I won't say it works great, but it's helped me a few times.


edit to add: I search the web at large for lyrics, but also search some keywords on https://setlist.FM and https://Discogs.Com, and browse shows on https://etreedb.org/ as well, of course, as the archive!

Use stems separation software to isolate the vocals.

[Gutbucket]

A lot of vocal energy is concentrated in the range from approximately 800 - 2000 cycles so I often start by cutting everything else.

Same, combined with boosting around 2.5 - 3k to enhance articulation of distant/buried vox.  Sort of a quick and easy manual filtering method emulating a rudimentary vox stem separation. 

I'm trying to recall the specific term describing that particular aspect of speech intelligibility. A quick search didn't help, but I was able to confirm that linguists who specialize in the physical properties of speech are called phoneticians.. not phoenicians.  "Fraught with pharmaceutical fatigue yet figuring he might fuck around with formant frequencies to finagle the faint falsetto from the frenetic fanfare, Phil the Phoenician phonetician PhD flopped down on the futon, feigned fortitude and, fumbling with his phone, fell into the flow and found a few phantom fricatives ferreted away in the fundamental frequencies.. figuratively at least."