A spot is carried either by voice or by music. In 47 Czech TV spots across 13 brands, 44 are voice-led, 3 are music-led, 0 mixed. Czech advertising diversifies how it looks, and converges on how it sounds.
Audio-only m4a downloads from the same 47 spots as the color study. Voice activity via silero-vad; tempo via librosa beat-tracking; key by Krumhansl-Schmuckler chroma correlation; spectral centroid and RMS dynamics via librosa. Voice-led is >50% speech across the spot. Music-led is >50% non-speech musical content. The middle is rare.
01
The split
One bar per spot. Black is voice, red is music, warm gray is silence. Sorted by music share descending. The almost-empty top end is the finding: voice-led is the default; music-led is the deliberate departure.
v / m percent shown right of each bar.
02
Per-brand audio identity
For each brand, the average voice / music split across its spots, plus tempo, mode majority, dynamic range, and brightness (spectral centroid).
Lidl
Retail · 5 spots
Albert
Retail · 5 spots
Komerční banka
Banking · 2 spots
Škoda Auto
Automotive · 5 spots
Vodafone
Telecom · 5 spots
Mattoni
Beverage · 2 spots
AAA Auto
Automotive · 1 spot
Kofola
Beverage · 1 spot
T-Mobile
Telecom · 5 spots
Fio banka
Banking · 5 spots
Penny Market
Retail · 5 spots
McDonald's
QSR · 5 spots
Pilsner Urquell
Beverage · 1 spot
03
The numerical defaults of Czech ad audio
Tempo lives in one band. Mode skews minor. Czech advertising audio does not vary the way the picture does.
Spot count per 10-bpm tempo bin. The 110-140 range absorbs 23 of 47 spots in the corpus.
Mode share across 47 spots
Minor mode dominates regardless of category. Even retail promos run in minor keys most of the time.
04
Brand similarity by sound
Hierarchical clustering on standardised audio features (voice / music balance, tempo, spectral centroid + bandwidth, RMS mean + std, onset rate). Two brands joining low share an audio signature. Pilsner Urquell branches alone; everyone else clusters in a tight family.
Euclidean distance after z-standardisation; UPGMA linkage.
05
Patterns
01
Voice is the default. Music-led is one brand.
Of 47 Czech TV spots from 13 brands, 44 run above 50% voice. Only three spots cross the music threshold: Pilsner Urquell's "Alfons" image spot (86% music), and two McDonald's image spots (77% and 76% music). After Pilsner the next cliff drops to 41%. There is no music-led promo spot in the corpus.
Why it matters"Music carries the spot" is treated as a creative variable in international advertising; in Czech advertising it is essentially unused. The cost to occupy that empty quadrant is low, the distinctiveness is automatic.
02
The numerical defaults of Czech ad audio are narrow.
Tempo clusters in 110-140 bpm (33 of 47 spots). Minor mode dominates (27 of 47). Spectral centroid sits in a tight 2000-2900 Hz band across categories. Dynamic range is flat: median RMS std 0.057, no ad in the corpus uses dramatic build-and-payoff loudness.
Why it mattersThe audio profile of Czech advertising is one playlist. A spot that breaks the tempo band, the mode default or the dynamic-range flatness will register as distinctive before a word is heard.
03
Per-category audio identity is sharper than per-brand identity.
Retail (Lidl 90% voice, Albert 88%, Penny 64%) lives at the voice end. Beverage and QSR allow more music (Mattoni 23%, Kofola 26%, McDonald's 40%). Banking and Automotive sit in between (Fio 31% music, Škoda 20%). The clusters by sector hold; brand-specific audio identity is the exception.
Why it mattersAn audio brief built around "what does your brand sound like" will produce category-defaults. An audio brief built around "what does your brand sound like compared to your category" forces the actual decisions.
04
Music-led spots are slow and rare.
The three music-led spots run 81, 70 and 70 bpm — well below the 127 bpm corpus median. Pilsner's spot is in F# major, both McDonald's image spots in C# minor. Music-led is associated with half-tempo cinematic pacing, not faster builds.
Why it mattersWhen a brief asks for "an emotional spot", the structural answer is: slow the tempo, let the music carry, drop the voiceover. The data shows that combination is reproducible and underused.
06
Playbook
Five moves a strategist or sound designer can run on next week's pitch.
01
Pick voice-led or music-led in the brief.
The corpus is bimodal: 44 voice-led, 3 music-led, almost no middle. The middle does not work because it cannot be reproduced; deciding one of the two camps gives the sound designer a clear target. Voice-led: target 80% speech, 120 bpm bed in a minor key. Music-led: target sub-30% speech, sub-90 bpm tempo, primary musical key.
02
If voice-led: the VO must carry meaning, because the music does not.
Lidl runs 90% voice with promotional copy throughout; the music under it is decoration, not signal. The cost of a flat VO is a forgettable spot. Cast for voice as carefully as for face.
03
If music-led: budget for a real sound bed, not a stock loop.
Pilsner's spot uses a single composed passage that reads as cinematic; stock loops at the same tempo do not. Music-led is the cheapest visual distinctiveness move in Czech advertising right now, but only when the music is itself distinctive.
04
Score competitors on the voice / music split as a pitch slide.
Same move as the color project. Show the client where their category currently sits on the audio axis. For most Czech categories that's "voice-led, 120 bpm, minor key, flat dynamics". Then name where their spot will land. Distinctiveness as a coordinate.
05
Score your audio mix on dailies, not at delivery.
The pipeline runs on rough audio in minutes. If the brief was music-led (target voice fraction under 30%) and the dailies are scoring 70%, the mix is drifting. Catching that in pre-mix is cheaper than re-recording VO after picture lock.