


ABOUT
—
ABOUT
—
ABOUT
—
I study where
speech loses
credibility
I study where
speech loses
credibility
I study where
speech loses
credibility
My work explores the fracture point between AI-generated Hungarian speech and human perception. The Illusion Breaks in Hungarian is a research project examining semantic loss, timing drift, dubbing quality and speech credibility.
Through dubbing reconstruction, sync correction and ASR analysis, I examine how timing, rhythm and semantic distortion alter meaning.
Focus: speech timing, dubbing QA, ASR/TTS evaluation and Hungarian linguistic review.
Accuracy alone is not communication
My work explores the fracture point between AI-generated Hungarian speech and human perception. The Illusion Breaks in Hungarian is a research project examining semantic loss, timing drift, dubbing quality and speech credibility.
Focus: speech timing, dubbing QA, ASR/TTS evaluation and Hungarian linguistic review.
Accuracy alone is not communication
My work explores the fracture point between AI-generated Hungarian speech and human perception. The Illusion Breaks in Hungarian is a research project examining semantic loss, timing drift, dubbing quality and speech credibility.
Focus: speech timing, dubbing QA, ASR/TTS evaluation and Hungarian linguistic review.
Accuracy alone is not communication.


— DUBBING / TTS / SYNC QUALITY LAB
— DUBBING / TTS / SYNC QUALITY LAB
— DUBBING / TTS / SYNC QUALITY LAB
We rebuild
what meaning has lost.
We rebuild
what meaning
has lost.
We rebuild
what meaning
has lost.
Neural dubbing, TTS evaluation and focused Hungarian adaptation for
systems where literal accuracy is not enough.
HEAR WHAT WAS MEANT →
Neural dubbing, TTS evaluation and focused
Hungarian adaptation for
systems where literal accuracy is not enough.
HEAR WHAT WAS MEANT →
Neural dubbing, TTS evaluation and focused
Hungarian adaptation for
systems where literal accuracy is not enough.
HEAR WHAT WAS MEANT →



Vowel length mismatch, late mouth opening
Corrected timing & duration

Phoneme collapse, unnatural transitions
Natural phoneme transitions

Breath misalignment, cut-off ending
Breath matched, natural ending


TIMING DRIFT
Time-stretch & realignment
LIP-SYNC ERROR
Phoneme-based re-synthesis
EMOTION MISMATCH
Re-performance & tone match
VOICE CONSISTENCY
Spectral matching
FILMHungarian Lip-Sync Reconstruction
Timing correction, phoneme alignment, semantic preservation.
TTSNeural Voice Evaluation
Hungarian prosody, intelligibility and naturalness testing.
QADubbing QA Error Atlas
Error classification for localization and dubbing pipelines.
FILMHungarian Lip-Sync Reconstruction
Timing correction, phoneme alignment, semantic preservation.
TTSNeural Voice Evaluation
Hungarian prosody, intelligibility and naturalness testing.
QADubbing QA Error Atlas
Error classification for localization and dubbing pipelines.
AUDIO —
AUDIO —
AUDIO —
Audio-Visual
Timing & Editing
Audio-Visual
Timing & Editing
Audio-Visual Timing & Editing
Editing rhythm, sound-image synchronization
and cinematic pacing.
Editing rhythm, sound-image synchronization
and cinematic pacing.
Editing rhythm, sound-image synchronization
and cinematic pacing.
15+ years of audiovisual work with young
classical musicians — recording, concert
documentation and promotional editing.
Focused on rhythm, pacing and perceptual
sync between sound and image.
15+ years of audiovisual work with young
classical musicians — recording, concert
documentation and promotional editing.
Focused on rhythm, pacing and perceptual
sync between sound and image.
15+ years of audiovisual work with young
classical musicians — recording, concert
documentation and promotional editing.
You either hear it or you don't
.
You either hear it or you don't
.
You either hear it or you don’t




CLINICAL ASR FAILURE
ATLAS
CLINICAL
ASR FAILURE
ATLAS
CLINICAL ASR FAILURE
ATLAS
MAPPING SEMANTIC RISK IN HUNGARIAN MEDICAL SPEECH RECOGNITION
MAPPING SEMANTIC RISK
IN HUNGARIAN MEDICAL
SPEECH RECOGNITION
MAPPING SEMANTIC RISK IN HUNGARIAN MEDICAL SPEECH RECOGNITION
300 curated ASR error examples
6 semantic failure classes
P1–P3 severity taxonomy
Public Hungarian medical speech
EXPLORE ATLAS →
FAILURE CLASS I
NUMERICAL
SEMANTIC COLLAPSE
FAILURE CLASS I
NUMERICAL
SEMANTIC COLLAPSE
FAILURE CLASS I
NUMERICAL
SEMANTIC COLLAPSE
SOURCE: "kettő-három"
ASR: "23"
SOURCE: "kettő-három"
ASR: "23"
A clinically essential drug identifier disintegrates during speech reconstruction, introducing potential treatment risk.
A clinically essential drug identifier disintegrates during speech reconstruction, introducing potential treatment risk.
A clinically essential drug identifier disintegrates during speech reconstruction, introducing potential treatment risk.
SEVERITY: P1
SEVERITY: P1





FAILURE CLASS V
NEGATION INVERSION
FAILURE CLASS V
NEGATION INVERSION
FAILURE CLASS V
NEGATION INVERSION
SOURCE: "vérrögoldót adnak"
ASR: "vérrögoldót NEM adnak"
SOURCE: "vérrögoldót adnak"
ASR: "vérrögoldót NEM adnak"
SOURCE: "vérrögoldót adnak"
ASR: "vérrögoldót NEM adnak"
Clinical intent inverted.
A single inserted negation reverses treatment logic completely.
Clinical intent inverted.
A single inserted negation reverses treatment logic completely.
Clinical intent inverted.
A single inserted negation reverses treatment logic completely.
SEVERITY: P1
SEVERITY: P1
FAILURE CLASS I
TERMINOLOGICAL
DISINTEGRATION
FAILURE CLASS I
TERMINOLOGICAL
DISINTEGRATION
FAILURE CLASS I
TERMINOLOGICAL
DISINTEGRATION
SOURCE: "Clopidogrel"
ASR: "klopi dogrel"
SOURCE: "Clopidogrel"
ASR: "klopi dogrel"
A clinically essential drug identifier disintegrates during speech reconstruction, introducing potential treatment risk.
A clinically essential drug identifier disintegrates during speech reconstruction, introducing potential treatment risk.
A clinically essential drug identifier disintegrates during speech reconstruction, introducing potential treatment risk.
SEVERITY: P1
SEVERITY: P1


THE EVIDENCE BASE
300
THE EVIDENCE BASE
300
THE EVIDENCE BASE
300
CLINICAL ASR FAILURES
CLINICAL ASR FAILURES
CLINICAL ASR FAILURES
P1–P3 SEVERITY TAXONOMY
P1–P3 SEVERITY TAXONOMY
P1–P3 SEVERITY TAXONOMY
HUNGARIAN MEDICAL MULTI-ASR AUDIT
HUNGARIAN MEDICAL MULTI-ASR AUDIT
HUNGARIAN MEDICAL MULTI-ASR AUDIT
300 CURATED ASR ERROR EXAMPLES
300 CURATED ASR ERROR EXAMPLES
300 CURATED ASR ERROR EXAMPLES
Manually curated archive of Hungarian medical ASR errors.
Manually curated archive of Hungarian medical ASR errors.
QA / THINKING
—
QA / THINKING
—
QA / THINKING —
I evaluate where
meaning survives
reconstruction
I evaluate where
meaning survives
reconstruction
I evaluate where meaning
survives reconstruction
My QA work focuses on native Hungarian perception and semantic stability in reconstructed speech systems.
Evaluation areas:
— semantic distortion: meaning shifted or lost during transcription
— linguistic failure: native Hungarian patterns broken or unnatural
— timing instability: speech rhythm disrupting comprehension
— cognitive correction burden: extra mental effort required to restore meaning
Accuracy is not understanding
My QA work focuses on native Hungarian perception and semantic stability in reconstructed speech systems.
Evaluation areas:
— semantic distortion: meaning shifted or lost during transcription
— linguistic failure: native Hungarian patterns broken or unnatural
— timing instability: speech rhythm disrupting comprehension
— cognitive correction burden: extra mental effort required to restore meaning
Accuracy is not understanding
My QA work focuses on native Hungarian perception and semantic stability in reconstructed speech systems.
Evaluation areas:
— semantic distortion: meaning shifted or lost during transcription
— linguistic failure: native Hungarian patterns broken or unnatural
— timing instability: speech rhythm disrupting comprehension
— cognitive correction burden: extra mental effort required to restore meaning
Accuracy is not understanding.


LET’S
RESTORE
MEANING.
Available for collaboration
and new opportunities.
- ♡Lip-Sync & Dubbing Synchronization
- ⌁Speech Timing Optimization
- ◎Dubbing QA Review
- ▧TTS Quality Evaluation
- ▤ASR Evaluation

LET’S
RESTORE
MEANING.
Hungarian AI dubbing, lip-sync, speech timing and QA review for generated voices that need to survive the human ear.