Investigation · Big Hero 6 (2014) · 21 Sep 2026

How Baymax trails off

Baymax "dies" twice in the film, and only one of the two is played in the voice. The low-battery collapse is a continuous slide: pitch, loudness and tempo fall together for about a second, ending below the character's normal floor. At the portal, where he actually goes, the voice does not degrade at all.

−19 st

pitch glide on "battery": 188 Hz to 63 Hz, a semitone every 52 ms

1.10 s

length of the glide; the word takes roughly twice its normal time

−13 dB

loudness fade across the same 1.1 s

0.985

r² of a straight line in semitones: the fall is a constant-rate slide, not a step

The scene: "Low battery"

Baymax, eyes half-closed, pointing at Hiro in the police station

Watch on YouTube · 2:48

The scene, as uploaded to YouTube. It opens in a new tab: the hub serves pages in a sandbox without an origin, and YouTube's player refuses to run there. The "Low battery" line is at the start; the measured glide is the second word.

Hiro and Baymax have just jumped out of a window; at the police station Baymax's eyes half-close and he announces the state. The line is two words and the second one is the whole effect. "Low" rises 8.5 semitones (151 → 246 Hz) over 0.45 s, a small upward lurch, then "battery" starts at 188 Hz and slides without a break to 63 Hz over 1.10 s while fading 13 dB. The endpoint matters: Baymax's normal register bottoms out at roughly 100 to 115 Hz (10th percentile of the ordinary lines), so the slide ends a good octave below anything he says when healthy. It sounds like a battery-powered toy running down because that is the shape of one: a tape-stop.

Fundamental frequency (dots, log scale, Hz) with loudness as the shaded area. Clip: fan upload, 16 kHz mono, 10 ms hop.

What the "drunk" state is, measured

Scott Adsit's direction, given halfway through production, was to make low-power Baymax "slurring his words and being a little playful and a little more emotional" (his words, CinemaBlend). The measurements say what "slurring" turned out to mean acoustically, and it is not what one would guess.

Line	Median F0	Range (p10 to p90)	Slope across phrase	Duration
Normal · "I will scan you now. Scan complete."	125 Hz	12.0 st	+3.8 st/s	3.6 s
Normal · "I cannot deactivate until you say…"	191 Hz	11.5 st	−0.3 st/s	4.9 s
Low battery · "Low battery."	169 Hz	20.0 st	−10.2 st/s	2.2 s
Low battery · "I'm healthcare, your personal Baymax companion"	172 Hz	11.3 st	−2.0 st/s	4.2 s
Low battery · "I will scan you now. Scan complete."	216 Hz	17.6 st	−6.6 st/s	3.0 s
Portal · "There is no time. Are you satisfied with your care?"	132 Hz	13.7 st	−0.1 st/s	3.9 s

He does not slow down. The same sentence, "I will scan you now. Scan complete.", takes 3.0 s drunk and 3.6 s sober. Tempo is not the signal.
He does not drop. Median pitch in the drunk lines is the same or higher than sober (172 to 216 Hz vs 125 to 191 Hz).
He gets melodic. Range widens from about 12 semitones to 18 to 20, and every phrase now tilts downward (−2 to −10 st/s) where the sober phrases are flat. The sober robot speaks in level steps; the drunk one sings and sags.

Same sentence, sober (top) and low-battery (bottom). Sober: plateaus, narrow band. Low-battery: higher peaks, longer falls inside each word, and the second phrase starts a fifth higher than the first.

The scene where he actually goes

Inside the portal, thrusters dead, Baymax asks the one question he needs answered before he can shut down. Nothing happens to the voice. "There is no time. Are you satisfied with your care?" sits at 132 Hz median, flat across the phrase, with the ends of "time" and "care" dipping to about 73 Hz: the bottom of his register, used as weight rather than decay. "Hiro, I will always be with you" is low, slow and level under the score. Then Hiro says the words, the rocket fist fires, and the deactivation is carried by the eyes dimming and the music, not by any treatment of the voice.

Portal farewell. Flat, low, intact. The quiet stretch from 1.0 to 2.0 s is the pause after "time".

So the film keeps two vocabularies apart. Loss of power is a glide: the machine audibly running down, comic, recoverable, and the audience is told it is recoverable by the sound itself (a toy that winds down can be wound up). Death is the voice at full integrity, low in the register, and then silence. Making the farewell voice degrade would have said "malfunction" when the scene means "choice".

How it was made, as far as the record goes

Adsit recorded alone; Disney describes a "robotic overlay" added to his voice (Disney press junket via TheSuburbanMom). The sober lines' stepped, narrow-band pitch is consistent with a performed monotone plus light processing rather than a vocoder.
Baymax's non-vocal sound (the squeak, the deflation after "I'm satisfied with my care") came from a squeaky exercise ball at Shannon Mills's desk at Skywalker Sound; David Fluhr, dialogue mixer: "We didn't want to be locked into one treatment that was just robot the whole time" (Disney Digital Studio Services).
Whether the "battery" glide was performed by Adsit and then time-stretched, or rendered in the mix as a varispeed ramp, is not on the record. The pitch drop (a factor of 3) exceeding the time stretch (roughly a factor of 2) is what a ramp that keeps accelerating would produce, so a varispeed on a performed glide is the likeliest reading; I cannot confirm it from a mixed clip with score underneath.

What this says for a voice that has to trail off

Running down is a slide, not a step. One continuous glide of 15 to 20 semitones over about a second, with loudness falling 10 to 15 dB on the same curve, ending below the voice's normal floor. A cut or a fade alone reads as a dropout; the glide reads as the machine.

Degraded is not slower. The impaired state keeps its tempo and widens its pitch; the audience hears the loss of control in the melody, not the clock.

Do not degrade the voice at death. A deliberate ending gets the intact voice, low and level, and then nothing. Degrade it and it says malfunction.

Method and limits

Sources: fan uploads of three scenes (introduction, low battery, portal), audio only, 16 kHz mono. Word timings from whisper.cpp small.en; pitch from a 40 ms autocorrelation tracker at 10 ms hop, 60 to 400 Hz, voicing threshold 0.5. All clips carry score and effects under the dialogue, so absolute loudness is relative to each clip and the portal lines are the least reliable. Fandom transcripts and TV Tropes were unreachable (Cloudflare and paywall). No prior work on this in the shared archive.