The Grammys are unequivocally the biggest night in the US music industry calendar. But recent controversies around the nomination process have meant this year’s awards have seen sweeping changes. These changes were designed to encourage greater diversity in nominations. In this blog, we’ll use the Musiio Tag AI to learn whether this has worked and whether our other technologies might be able to help us predict winners in the Record of the Year and Best Engineered Album categories.
First up, we’re looking at the Record of the Year category to see whether we can see any more diversity versus the past three years. We’ve counted the number of different genre tags for each year’s nominations. In theory, if there’s a wider spread of music, we’d expect to see a higher number of genre tags for 2022 than in previous years. We calculate this number using data from our AI, which assigns up to four genres per track.
We see an increase in the number of genres in 2022, even if it’s only marginal. This change could be down to a couple of things. First is the higher number of nominations in the category this year – up from eight to ten. Second, the swing record ‘I Get A Kick Out Of You’, performed by Tony Bennett and Lady Gaga, is an outlier compared to previous years, driving up the number of detectable genres. Still, so far, so good, as far as the AI can tell.
People are calling this year’s Grammys a wide-open race because of the new criteria for voting members of the Recording Academy. For example, most have to be actively making music, meaning that 2022 could be the most culturally relevant Grammys ever. But what does that mean when it comes to predicting a winner?
We’re going to try and use the Musiio AI to make some predictions. Following up next week, we’ll see whether they turned out to be right. We’ll attempt to make selections in two categories based on the previous three years of data.
Can the ‘Quality’ tag measure engineering skill?
The first category is the Best Engineered Record, Non-Classical. For this category, we’ve taken the nominees since 2019 to see what trends we can see. Musiio’s Quality tag shows the AI’s assessment of a recording’s sound quality.
So, in theory, predicting the best Best Engineered Album should be as simple as feeding in the audio from those records and seeing which has the highest average score. The one caveat is that instead of processing audio from every track on every nominated album for the last three years, we picked out the top three performing songs on each album. We can see in this graph how our Quality tag correlates to Grammy wins in the Best Engineered, Non-classical category over the years.
This is fascinating. There’s not just zero correlation in the last three years, but in 2020, there was anti-correlation. Billie Eilish’s When We All Fall Asleep, Where Do We Go? won the category that year but had the lowest Quality score from our AI.
So what’s going on here? The Quality score – usually presented as named ranges from Very Low to Very High – can be useful for talent scouts weeding out low-quality live recordings or catalogue owners looking for pristine recordings for sync. But it doesn’t work that way for awards. Everything is already at a professional standard; this is the cream of the crop.
The records that score highest for Quality tend to be polished pop records, which could partly be down to the training data. In 2019, Charlie Puth’s wildly successful pop album Voicenotes scored highest with our AI, but the award went to the far quirkier, rockier and lower-rated Colors by Beck.
Since albums nominated in this category have already passed a quality threshold, lower scores may indicate more unusual recording styles or arrangements. That was certainly the case with the 2020’s winner, Billie Eilish’s brother Finneas O’Connell.
In 2022, The Marías’ Cinema has the highest Quality rating with its Eilish-ASMR-inspired pop perfection. However, Hey What by Low – the lowest-ranked for Quality – could easily win this year with its seasoned, articulate left-field sound. We won’t nail it down to one – we believe it could go either way.
Record of the Year prediction
The other category we’ll look at is the big-ticket Record of the Year. We can judge the nominees using our proprietary Hit Potential algorithm, which predicts a song’s potential commercial success based on audio alone.
Records with a commercially successful sound and records that win over judges clearly aren’t the same thing, though. Intriguingly, we see the same effect in the Record of the Year category as with the Best Engineered Album category. We don’t see the same level of anti-correlation, but based on historical data, the best strategy for predicting a winner could be choosing something with a lower Hit Potential score.
If we were to go for the track with the highest Hit Potential, ‘Peaches’ by Justin Bieber would be the clear winner. However, if we consider historical data, Billie Eilish’s ‘Happier Than Ever’ might be the safest bet – she has award-winning form, and her track has a lower Hit Potential. Beyond the AI’s Hit Potential score, the rest is interpretation, and the changes to voting rules mean we’re in a brave new Grammy world.
The field is certainly strong, including ‘Montero’, ‘drivers license’ and ‘Leave The Door Open’, and we can’t wait to see who snags the coveted trophy.
Is that a cop-out? Maybe, but we never set out to use this tech to predict award winners, so we’re hesitant to pin our reputation to any one outcome. That said, the unexpected anti-correlation in the Best Engineered Album category might be one to watch out for.
In next week’s blog, we’ll compare our results against the actual winners and look at how AI could help judges decide how to categorise genre-blended tracks.