Accent Removal and Accent Installation in Discrete Unit Space

Anonymous submission — ICASSP 2027

Author One1, Author Two1, Author Three2
1Affiliation One, 2Affiliation Two

Abstract

[PLACEHOLDER — replace with the submitted abstract.]

One or two sentences on the setting: textless accent conversion in discrete unit space with real parallel supervision, over the bidirectional accent map (L2 → L1 removal and L1 → L2 installation).

One or two sentences on the finding: conversion splits into removal and installation, the two are asymmetric, and the asymmetry is a property of the substrate rather than of the model.

One sentence on what the samples below show, and one on the evaluation.

Audio samples

Every row is one utterance, and every system on that row converted the same recording of it. Source and Target are natural recordings of the same prompt by the two speakers — the target is what a perfect conversion would sound like, not a system output. Ours and the baselines follow.

The 60 utterances here (30 removal, 30 installation) are exactly the stimuli of the subjective listening test, chosen by a blind triage pass in which items were judged unlabelled and shuffled.

1. Accent removal L2 → L1

Non-native source, native target. The conversion must remove the source speaker's accent while keeping the content.

Arabic → Native (L1)L2 → L1

UtteranceSource (L2)Target (L1, natural)OursTokANCosyAccentRWAVevo-Style
"Knowing him I review the old Scandinavian myths with clearer understanding"
YBAA → clb · arctic_b0403
"Mops sir eagerly answered the sailor at the wheel"
YBAA → bdl · arctic_b0445
"New idea he volunteered brand new idea"
YBAA → bdl · arctic_b0525
"Eggshell is not good to eat"
YBAA → bdl · arctic_b0517
"But already he had composed himself"
ABA → bdl · arctic_b0515
"And each year something happened and I did not go"
YBAA → bdl · arctic_b0405

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Hindi → Native (L1)L2 → L1

UtteranceSource (L2)Target (L1, natural)OursTokANCosyAccentRWAVevo-Style
"The scents of strange vegetation blew off the tropic land"
ASI → clb · arctic_b0503
"The dirk mentioned by Wolf Larsen rested in its sheath on my hip"
RRBI → bdl · arctic_b0409
"Not a wheel moved in his empire"
ASI → bdl · arctic_b0475
"But already he had composed himself"
SVBI → bdl · arctic_b0515
"Isaac Ford the austere soldier of the Lord the old hypocrite"
SVBI → clb · arctic_b0453
"O'Brien emitted a shriek that sank swiftly to a gurgling sob"
SVBI → bdl · arctic_b0447

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Korean → Native (L1)L2 → L1

UtteranceSource (L2)Target (L1, natural)OursTokANCosyAccentRWAVevo-Style
"He plodded on for half an hour when the hallucination arose again"
YDCK → bdl · arctic_b0513
"Also she wouldn't walk"
HKK → clb · arctic_b0449
"Well did they eat"
YDCK → bdl · arctic_b0435
"But already he had composed himself"
YDCK → bdl · arctic_b0515
"And he thought of Oona and of her words"
HKK → clb · arctic_b0521
"The dirk mentioned by Wolf Larsen rested in its sheath on my hip"
HKK → bdl · arctic_b0409

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Mandarin → Native (L1)L2 → L1

UtteranceSource (L2)Target (L1, natural)OursTokANCosyAccentRWAVevo-Style
"Do you value your hide"
LXC → bdl · arctic_b0511
"The Russian music player the Count was her obedient slave"
LXC → clb · arctic_b0523
"Not a wheel moved in his empire"
BWC → bdl · arctic_b0475
"But already he had composed himself"
NCC → bdl · arctic_b0515
"You used to joy ride like the very devil"
NCC → clb · arctic_b0451
"Of course much grumbling went on and little outbursts were continually occurring"
BWC → bdl · arctic_b0407

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Spanish → Native (L1)L2 → L1

UtteranceSource (L2)Target (L1, natural)OursTokANCosyAccentRWAVevo-Style
"The scents of strange vegetation blew off the tropic land"
EBVS → bdl · arctic_b0503
"After all superfluous flesh is gone what is left is stringy and resistant"
ERMS → bdl · arctic_b0487
"The task we set ourselves was threefold"
ERMS → clb · arctic_b0481
"Knowing him I review the old Scandinavian myths with clearer understanding"
MBMPS → bdl · arctic_b0403
"It was sanctification and salvation"
ERMS → bdl · arctic_b0457
"The task we set ourselves was threefold"
MBMPS → clb · arctic_b0481

The utterance column stays fixed — scroll the table sideways on a narrow screen.

2. Accent installation L1 → L2

Native source, non-native target. The conversion must install the target accent — the harder of the two directions, and the one the paper analyses.

Native (L1) → HindiL1 → L2

UtteranceSource (L1)Target (L2, natural)OursTokAN L1→L2RWAVevo-Style
"He gave no reason but his motive was obvious enough"
bdl → RRBI · arctic_b0411
"The Russian music player the Count was her obedient slave"
clb → RRBI · arctic_b0523
"New idea he volunteered brand new idea"
bdl → SVBI · arctic_b0525
"And there was Ethel Baird whom also you must remember"
clb → RRBI · arctic_b0441
"He gave no reason but his motive was obvious enough"
bdl → SVBI · arctic_b0411
"How valiantly I went at it that first day"
bdl → RRBI · arctic_b0505

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Native (L1) → KoreanL1 → L2

UtteranceSource (L1)Target (L2, natural)OursTokAN L1→L2RWAVevo-Style
"Mops sir eagerly answered the sailor at the wheel"
bdl → HKK · arctic_b0445
"There were orange green gold green and a copper green"
bdl → YDCK · arctic_b0425
"He gave no reason but his motive was obvious enough"
clb → HKK · arctic_b0411
"Your father's fifth command he nodded"
clb → HKK · arctic_b0419
"Not a wheel moved in his empire"
clb → YDCK · arctic_b0475
"Also she wouldn't walk"
clb → YDCK · arctic_b0449

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Native (L1) → MandarinL1 → L2

UtteranceSource (L1)Target (L2, natural)OursTokAN L1→L2RWAVevo-Style
"We are cooking on the coal stove and on the oil burners"
clb → LXC · arctic_b0433
"Of course much grumbling went on and little outbursts were continually occurring"
bdl → BWC · arctic_b0407
"You were making them talk shop Ruth charged him"
clb → BWC · arctic_b0539
"Let us run them for ourselves"
clb → BWC · arctic_b0467
"He plodded on for half an hour when the hallucination arose again"
clb → LXC · arctic_b0513
"The next thing to watch out for is bed sores"
bdl → BWC · arctic_b0417

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Native (L1) → SpanishL1 → L2

UtteranceSource (L1)Target (L2, natural)OursTokAN L1→L2RWAVevo-Style
"New idea he volunteered brand new idea"
clb → ERMS · arctic_b0525
"Do you value your hide"
clb → MBMPS · arctic_b0511
"The task we set ourselves was threefold"
clb → MBMPS · arctic_b0481
"He had fulfilled his duty and paid properly"
bdl → ERMS · arctic_b0509
"I said and dismissed the matter as not worth thinking about"
bdl → ERMS · arctic_b0499
"But such divergence of opinion would constitute no menace to society"
clb → ERMS · arctic_b0465

The utterance column stays fixed — scroll the table sideways on a narrow screen.

Native (L1) → VietnameseL1 → L2

UtteranceSource (L1)Target (L2, natural)OursTokAN L1→L2RWAVevo-Style
"He had fulfilled his duty and paid properly"
bdl → THV · arctic_b0509
"He had fulfilled his duty and paid properly"
bdl → HQTV · arctic_b0509
"Then there was the campaign"
clb → PNV · arctic_b0473
"There were orange green gold green and a copper green"
bdl → HQTV · arctic_b0425
"Then there was the campaign"
clb → HQTV · arctic_b0473
"Isaac Ford the austere soldier of the Lord the old hypocrite"
clb → THV · arctic_b0453

The utterance column stays fixed — scroll the table sideways on a narrow screen.

BibTeX

@inproceedings{placeholder2027accent,
  title     = {Accent Removal and Accent Installation in Discrete Unit Space},
  author    = {Anonymous},
  booktitle = {ICASSP},
  year      = {2027}
}