Category: research

News
FA-Bench Leaderboard for ASR with Word Timestamps
Research

FA-Bench Leaderboard for ASR with Word Timestamps

We ranked 28 speech recognition pipelines, 19 built from open models and 9 that call a commercial API, by how closely their word timestamps land on the boundarโ€ฆ

10์›” 5, 2026ยท 4 min read
FA-Bench, a Benchmark for the Accuracy of Word and Phone Timestamps
Research

FA-Bench, ๋‹จ์–ด์™€ ์Œ์†Œ ํƒ€์ž„์Šคํƒฌํ”„์˜ ์ •ํ™•๋„๋ฅผ ์žฌ๋Š” ๋ฒค์น˜๋งˆํฌ

FA-Bench๋Š” ๋‹จ์–ด์™€ ์Œ์†Œ ํƒ€์ž„์Šคํƒฌํ”„๊ฐ€ ์–ธ์–ดํ•™์ž๊ฐ€ ์ง์ ‘ ํ‘œ์‹œํ•œ ๊ฒฝ๊ณ„์— ์–ผ๋งˆ๋‚˜ ๊ฐ€๊น๊ฒŒ ๋†“์ด๋Š”์ง€ ์ธก์ •ํ•ฉ๋‹ˆ๋‹ค. TIMIT๊ณผ Buckeye์—์„œ ASR ์‹œ์Šคํ…œ๊ณผ forced aligner์˜ ์ ์ˆ˜๋ฅผ ๋งค๊ธฐ๋ฉฐ, ๊นจ๋—ํ•œ ์Œ์„ฑ๊ณผ ์žก์Œ์„ ๋”ํ•œ ์Œ์„ฑ์„ ๋ชจ๋‘ ๋‹ค๋ฃน๋‹ˆ๋‹ค. ๋ฆฌ๋”๋ณด๋“œ๋Š” 2026๋…„ 10์›” ๊ฒฐ๊ณผ๋กœโ€ฆ

8์›” 17, 2026ยท 4 min read
What Makes Olewave's Speech Data Cleaning Pipeline Effective and Unique
Research

Olewave์˜ ์Œ์„ฑ ๋ฐ์ดํ„ฐ ์ฒญ์†Œ ํŒŒ์ดํ”„๋ผ์ธ์„ ํšจ๊ณผ์ ์ด๊ณ  ๋…ํŠนํ•˜๊ฒŒ ๋งŒ๋“œ๋Š” ๊ฒƒ์€ ๋ฌด์—‡์ธ๊ฐ€

Olewave์˜ Olign ํŒŒ์ดํ”„๋ผ์ธ์ด ์›๋ณธ ์˜ค๋””์˜ค๋ฅผ ์‹ค์ œ ์„œ๋น„์Šค์— ๋ฐ”๋กœ ์“ธ ์ˆ˜ ์žˆ๋Š” ASR/TTS ํ•™์Šต ๋ฐ์ดํ„ฐ๋กœ ๋ฐ”๊พธ๋Š” ๋ฐฉ๋ฒ•์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ์—๋Š” ๊ฒ€์ฆ๋œ ์ „์‚ฌ๋ฌธ, ๋‹จ์–ด ๋‹จ์œ„ ํƒ€์ž„์Šคํƒฌํ”„, ์‹ ๋ขฐ๋„ ์ ์ˆ˜๊ฐ€ ํฌํ•จ๋˜๋ฉฐ, ์‚ฌ๋žŒ ๋ ˆ์ด๋ธ”๋Ÿฌ๋Š” ํ•„์š”ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.

3์›” 9, 2025ยท 4 min read
Harnessing Wild Data for High-ROI Speech R&D โ€” Our ICASSP 2024 Spotlight
Events

๋†’์€ ROI๋ฅผ ์œ„ํ•œ ์•ผ์ƒ ๋ฐ์ดํ„ฐ ํ™œ์šฉ ์Œ์„ฑ R&D โ€” ๋‹น์‚ฌ์˜ ICASSP 2024 ์ดˆ๋Œ€ ์—ฐ์„ค

๋‹น์‚ฌ ์„ค๋ฆฝ์ž์ด์ž CEO์ธ Wei Chu์˜ ICASSP 2024 ์ดˆ๋Œ€ ์—ฐ์„ค โ€” ๊ธฐ์กด ์Œ์„ฑ ์„œ๋น„์Šค ๊ตฌ์ถ•์ด ๋‚ฎ์€ ROI์ธ ์ด์œ ์™€ ์•ผ์ƒ ๋ฐ์ดํ„ฐ ํ•„ํ„ฐ๋ง + ๋Œ€ํ˜• ์˜คํ”ˆ์†Œ์Šค ์Œ์„ฑ ๋ชจ๋ธ๋กœ ๋น„๊ณต๊ฐœ ๋ฐ์ดํ„ฐ ์ž๋™ ๋ ˆ์ด๋ธ”๋ง์ด ๋ฐฉ์ •์‹์„ ์–ด๋–ป๊ฒŒ ๋’ค์ง‘๋Š”์ง€์— ๋Œ€ํ•ด

4์›” 16, 2024ยท 2 min read