My favorite band sings in Finnish.

Maustetytöt are an indie-pop/iskelmä duo. Their name means "spice girls," like the British band, an inside joke. They sing about depressing topics but in a way that is surprisingly uplifting. Sort of Morrissey meets the Magnetic Fields (meets Markus Allan).

Despite singing in Finnish, they have gained an international following (especially in Europe) after appearing in the excellent Aki Kaurismäki film Fallen Leaves.

They were subtitled in the film and I suspect owe much of their extra-Finland popularity -- they sold out venues in Paris and Berlin in 2024 -- to the existence of Lyrcis Translate (the website) and an English subtitled, and very well shot and perfomed, set of songs from a concert at Helsinki Kulttuutitalo in late 2020.

I don't speak Finnish, but by the time I came across Maustetytöt they already had three albums out and every song had English lyrics available somewhere on the internet. And even early songs that are rarely heard any more had Finnish lyrics that could be machine translated with a click (except this one, which a kind Finnish soul transcribed after I begged in the comments.... kiitos "lilli m").

But yesterday Maustetytöt released the first single, "Oslo Helsinki Berliini Kokkola," from their forthcoming album. When I put on my headphones last night, I found it had a captivating, restrained sound, brighter and more spare than their past productions, and was clearly about the band's relentless 2023-2024 tour schedule, where they hit 96 cities with 166 shows (versus Taylor Swift's 53 and 166 on her tour, in almost precisely the same date range).

But how to actually understand what they were saying? There were no native Finnish lyrics up yet, much less English translations.

I turned to the same tool I use to transcribe recorded work interviews, OpenAI's Whisper. I use a packaged version of whisper for the Mac. I give it an audio file and it gives me an interactive transcript of my interview. I always double or triple check its output against the original audio before publishing anyone's words, and have found it to need shockingly few correctoins.

Until a few days ao I didn't realize Whisper (at least the large model) can also transcribe singing. Boy, can it.

After downloading the YouTube version of the new Maustetytöt single and reducing it to audio, I handed the file to Whisper which seemed to produce a transcript that seemed plausible to a non-Finnish speaker. The words seemed to line up with what I heard on the audio (one nice thing about Finnish is that pronounciation of letters remains consistent across words, unlike in English). I could see there were some errors even without understanding the transcription; for example words changing from one iteration of the chorus to another even though they sounded the same on the recording. But generally the accuracy seemed prety good.

Then I translated the Finnish text using Google Translate. Whisper can translate directly but my copy, MacWhisper, for some reason does not use Whisper's translation capability, instead falling back on Apple's on-computer translation system, which cannot handle Finnish.

The translation mostly seemed to fit the theme of the song. I did some spot checks on lines that seemed to not make sense or in context or were otherwise jarring. Correcting those lines involved a fairly desperate and scattershot attempt to understand them in a language I don't speak: Machine translating various fragments of a line or set of lines; searching phases; looking up individual words and their many, many Finnish declensions; asking Google search's built in AI (often on accident) about idioms or phrase meanings (surprisingly useful).

I had a pretty decent set of lyrics within about 20 minutes of deciding to put them together. The hardest part was formatting - putting Finnish lines next to English ones.

I've been fixing little things ever since. For example, a news story (in Finland) about the new single said it mentioned Juhla Mokka, the leading coffee bean brand in Finland. This mad me realize Whisper had misheard this as "juhlamakasta," which roughly means party (so Google Translate tells me). I had thought the singer said "I like to party / I don't like the spotlight." Actually it was "I like Juhla Mokka [staying home with coffee] / I don't like the spotlight", which makes a lot more sense!

The next day someone (of course) wrote the Finnish lyrics into a comment under the YouTube video. It looks like Whisper did a very good job, matching the human transcription almost verbatim.

Google Translate seems to have fallen down a bit more, but it at least let me accurately understand the bulk of what I was hearing. So much of the experience of a rock song is not about the words. Even the experience of listening to a singer is as much about the sound of their voice as the words they say. If you can add a 80-90% accurate translation to that you're not missing tooo much. Especially when time is of the essence; on the internet, a better translation will come in time.

You can find my transcript and (mostly machine) translation of Maustetytöt's new single Oslo Helsinki Berliini Kokkola here. It's likely still got some errors but I'll probably keep finding and fixing some of them over the next few days at least.

You can find all my English language Maustetytöt lyrics here. That page only contains things I could not find native translations for. The rest of the band's songs are in English (via actual Finnish speakers) on Lyrics Translate. Nothing beats human translation -- when you can get it.