01 · THE VOICE
+182.9%
Familiar sounds 182.9% more like the real speaker0.399 vs 0.141 speaker resemblance · 31 exactly aligned dubs
THEIR AUTO-DUB SPEAKS IN A STRANGER'S VOICEFAMILIAR VS YOUTUBE AUTO-DUBBING · 50 DUB PAIRS · 5 LANGUAGES · AUGUST 2026
Tested more accurate than YouTube auto-dubbing.
01 · THE VOICE
+182.9%
02 · THE MEANING
−70.3%
Familiar made 70.3% fewer translation errors; changed or lost important ideas43 vs 145 of 528 important source ideas · 50 whole dubs reviewed
CRITICAL MEANING FAILURES: YOUTUBE 12 · FAMILIAR 0
03 · THE LAUGHS
0.998
Laughter and reactions keep their shapevs 0.236 shape correlation · 14 event lanes · loudness error 0.11 vs 3.36 dB
04 · THE SCENE
Keeps the music, the ambience, the scene under the dub.
+232.6% more background-sound error · YouTubethe noise bed: music, effects, and room ambience · same clips
A usable YouTube Auto-Dub existed for 25 of 43 source clips — Familiar dubbed all 43
Frozen cohort · captured August 8, 2026 · complete delivered final audio only
SIX SYNCHRONIZED EXAMPLES · ORIGINAL · FAMILIAR · YOUTUBE AUTO-DUB
Hear the difference
FROZEN COHORT · EVERY METRIC WITH ITS OWN ELIGIBLE LANES
Final-audio results: Familiar vs YouTube Auto-Dub
50 SELECTED DUB PAIRS · ARABIC · SPANISH · FRENCH · HINDI · KOREAN
Nine measures, one cohort
Paired within source clip and language; 95% intervals resample source clips.
- SIM-o: source vs dubbed speaker embeddings, eligible turns only
- higher = the dub still sounds like the same person
- 528 important spoken ideas reviewed across all 50 dubs
- an idea fails only on a major or critical meaning failure
- paraphrases + tiny reactions never count · CER/WER excluded
- critical = the review's most severe meaning grade
- dubs with any important issue: YouTube 34/50 · Familiar 18/50
- spectral error in verified speech-free regions: music, ambience, the scene
- mean absolute error in dB, lower is better · Familiar 69.9% lower
- the loudness envelope of each verified laugh or reaction, correlated vs the source
- 1.0 = rises and falls exactly like the original
- the level miss on the same verified event regions
- how far each laugh lands from the source level
- missed + extra sound-effect and beat onsets, balanced as F1
- Familiar kept every onset · reported descriptive: the interval touches zero
- model-predicted listening score · a diagnostic, not a human study
- the interval is broad and crosses zero
- a usable YouTube Auto-Dub existed for 25 of 43 catalog source clips · Familiar all 43
- the other 18 are reported unavailable, never scored as zero
METHOD + SOURCES · FROZEN ARTIFACT youtube-autodub-20260808-bb774ab9
How it was measured
Complete delivered final audio from both systems; nothing partial.
Decode, channel conversion, resampling, one constant trim. No time warping, gain changes, denoising, or padding; uncertain lanes fail closed.
Each metric uses only its eligible lanes; the 50-lane selection is context, not a universal sample size.
| # | Clip | Source + window | Length |
|---|---|---|---|
| 01 | This Woman Made Up A Job!!😂😂Kill Tony ft.Jimmy Carr | 0:00–0:49 ↗ | 49s |
| 02 | Ali Abdaal - 01 - 00h03m16s - Why You’re Not Financially Free (Yet) | 3:16–4:16 ↗ | 60s |
| 03 | Ali Abdaal - 02 - 00h14m28s - i am begging you to try harry potter fan fiction | 14:28–15:28 ↗ | 60s |
| 04 | Binging with Babish - 01 - 00h07m36s - I Made EVERY Salsa | 7:36–8:36 ↗ | 60s |
| 05 | Binging with Babish - 02 - 00h11m04s - Recreating Bobby Hill’s Restaurant Menu ｜ Binging with Babish | 11:04–12:04 ↗ | 60s |
| 06 | Botez Sisters - 00m20s-00m35s - Can I Beat Will Smith in 60 Seconds？ | 0:20–0:35 ↗ | 15s |
| 07 | BotezLive - 01 - 00h04m48s - Japan Changed Us... ｜ Botez Sisters ASIA VLOG | 4:48–5:48 ↗ | 60s |
| 08 | BotezLive - 02 - 00h07m59s - What ACTUALLY Happened At Mr Beast... | 7:59–8:59 ↗ | 60s |
| 09 | Bryce Crawford - 01 - 00h11m19s - Debating A “Gay” Pastor! | 11:19–12:19 ↗ | 60s |
| 10 | Bryce Crawford - 02 - 00h21m03s - Preaching the Gospel at the Pride Parade! | 21:03–22:03 ↗ | 60s |
| 11 | Ethan - 01 - 00h04m38s - I Ate Every Ramen (so you don't have to) | 4:38–5:38 ↗ | 60s |
| 12 | Ethan - 02 - 00h21m50s - I Flew to Japan Just for This | 21:50–22:50 ↗ | 60s |
| 13 | George Janko - 01 - 00h18m51s - HE TALKED TRASH—THEN FOUGHT MY BODYGUARD ｜ EP. 165 | 18:51–19:51 ↗ | 60s |
| 14 | George Janko - 02 - 01h02m01s - Reacting to Andrew Tate’s Arrest ｜ EP. 164 | 1:02:01–1:03:01 ↗ | 60s |
| 15 | Jacksfilms - 01 - 00h07m48s - Jack Douglass but the questions get Jack Doug-less good | 7:48–8:48 ↗ | 60s |
| 16 | Jacksfilms - 02 - 00h22m16s - Anthony Padilla but the questions get Anthony Badilla | 22:16–23:16 ↗ | 60s |
| 17 | Jay Shetty Podcast - 01 - 00h04m11s - High Achievers Manifest Using Science NOT Luck ｜ Emily McDonald | 4:11–5:11 ↗ | 60s |
| 18 | Jay Shetty Podcast - 02 - 00h12m13s - #1 Brain Doctor： Do THIS The Moment You Wake Up (Most Get It Wrong) ｜ Dr. Daniel Amen | 12:13–13:13 ↗ | 60s |
| 19 | Jeff Nippard - 01 - 00h08m17s - How Ripped Can I Get My Brother In 100 Days？ | 8:17–9:17 ↗ | 60s |
| 20 | Jeff Nippard - 02 - 00h08m10s - 7 Things That Helped My Shoulders Grow | 8:10–9:10 ↗ | 60s |
| 21 | Johnny Chang - 01 - 00h04m30s - Why You Need To Run To Christ | 4:30–5:30 ↗ | 60s |
| 22 | Johnny Chang - 02 - 00h15m39s - Destroy The Old You | 15:39–16:39 ↗ | 60s |
| 23 | Kallaway Marketing - 01 - 00h05m56s - Everything I Learned Growing From 0 to 2M+ Followers in 3 Years | 5:56–6:56 ↗ | 60s |
| 24 | Kallaway Marketing - 02 - 00h13m28s - 5 Claude Skills To Grow Faster Than 99% of People on Social Media | 13:28–14:28 ↗ | 60s |
| 25 | Larry Wheels - 01 - 00h04m39s - Is this the MOST dangerous Martial Art？ | 4:39–5:39 ↗ | 60s |
| 26 | Larry Wheels - 02 - 00h10m51s - The Biggest Bodybuilder You Never Heard of | 10:51–11:51 ↗ | 60s |
| 27 | Podcast - 00m20s-00m48s - Michelle Obama： We Still Go High | 0:20–0:48 ↗ | 28s |
| 28 | The Diary Of A CEO - 01 - 00h19m51s - Ray Dalio： I Predicted The 2008 CRASH, I Know What Comes Next! | 19:51–20:51 ↗ | 60s |
| 29 | The Diary Of A CEO - 02 - 01h09m47s - Vitamin D Expert： The Supplement World Is Giving The WRONG Advice! | 1:09:47–1:10:47 ↗ | 60s |
| 30 | Xander Budnick - 01 - 00h10m08s - 5 Days Alone with the Wilderness | 10:08–11:08 ↗ | 60s |
| 31 | Xander Budnick - 02 - 00h17m48s - I Climbed to the Highest Point on Earth (Higher than Everest) | 17:48–18:48 ↗ | 60s |
| 32 | xQc - 01 - 00h07m45s - Tony Romo Mouths Off to Cops in OWI Arrest ｜ xQc Reacts | 7:45–8:45 ↗ | 60s |
| 33 | xQc - 02 - 00h24m21s - xQc Reacts to the FUNNIEST TikToks & Videos From Chat! | 24:21–25:21 ↗ | 60s |
| 34 | 001 I ＂Cheated＂ Until the Arcade Banned Me - Early 01m55s | 1:55–2:55 ↗ | 60s |
| 35 | 003 01 Cooking Pizza with Matt Damon and Tom Holland - Early 01m42s | 1:42–2:42 ↗ | 60s |
| 39 | Living Raised with Christ | Ps Sam Picken | 29:03–34:03 ↗ | 300s |
| 44 | Triangle sound test | 0:00–1:09 ↗ | 69s |
| 45 | MTS - Teams Skipping Frontier-Model Security Scans Are Delinquent | 0:00–1:25 · non-YouTube source | 85s |
| 46 | Jimmy Yang 10s | 3:47–3:58 ↗ | 11s |
| 49 | Paul Graham - Why AI Can't Replace Founders | 54:23–55:15 ↗ | 52s |
| 50 | Lindsay DeFranco - Running for Georgia State House District 47 | 0:12–3:09 ↗ | 177s |
| 51 | Frozen - Elsa Flees From Arendelle Clip (HD) | 0:00–3:30 ↗ | 210s |
| 52 | Frozen - Meeting Olaf Clip (HD) | 0:00–2:17 ↗ | 137s |
| 53 | Shrek 2 - I Need a Hero - Jennifer Saunders | 0:00–4:37 ↗ | 277s |
A frozen black-box study of the captured cohort — not a claim about every video, language, region, or future platform version. Captured August 8, 2026.