πŸ› οΈAI 도ꡬ2026-06-08

λ‰΄μŠ€ - 원문 기반 μš”μ•½ ν•„μš”

πŸ’‘ ν•œμ€„ μš”μ•½|λ‰΄μŠ€ - 원문 기반 μš”μ•½ ν•„μš”


title: "0.4μ΄ˆλ§ˆλ‹€ 말할지 κ²°μ •ν•˜λŠ” μ˜€ν”ˆμ†ŒμŠ€ μŒμ„± AI λ“±μž₯" description: "λ‰΄μŠ€ - 원문 기반 μš”μ•½ ν•„μš”" date: 2026-06-08 tags: [ai-news] source: "https://the-decoder.com/new-open-source-voice-model-listens-nonstop-and-decides-every-0-4-seconds-whether-to-speak-or-stay-silent/" sidebar: order: 0

제λͺ©(ν•œκΈ€): 0.4μ΄ˆλ§ˆλ‹€ 말할지 κ²°μ •ν•˜λŠ” μ˜€ν”ˆμ†ŒμŠ€ μŒμ„± AI λ“±μž₯ 원문 제λͺ©(영문): New open-source voice model listens nonstop and decides every 0.4 seconds whether to speak or stay silent 원문: New open-source voice model listens nonstop and decides every 0.4 seconds whether to speak or stay silent μ†ŒμŠ€: the-decoder MD 파일: content/2026-06-08/the-decoder-new-open-source-voice-model-listens-nonstop-and-de.md

핡심 λ‚΄μš©

μ˜€λ””μ˜€ μŠ€νŠΈλ¦Όμ„ λŠκΉ€ 없이 λ“£λ‹€κ°€ 0.4μ΄ˆλ§ˆλ‹€ 'λ§ν• κΉŒ, λ“€μ„κΉŒ'λ₯Ό 슀슀둜 νŒλ‹¨ν•˜λŠ” μŒμ„± AIκ°€ κ³΅κ°œλμ–΄μš”.

κΈ°μ‘΄ GPT-4oλ‚˜ Qwen 같은 λͺ¨λΈμ€ λ…ΉμŒμ΄ λλ‚˜μ•Ό μ‘λ‹΅ν–ˆκ±°λ“ μš”. Audio-Interaction은 슀트리밍 도쀑에도 λ§€ 0.4초 μ²­ν¬λ§ˆλ‹€ / 토큰 ν•˜λ‚˜λ‘œ λ°˜μ‘ μ—¬λΆ€λ₯Ό κ²°μ •ν•΄μš”. λ²ˆμ—­Β·μ „μ‚¬Β·λŒ€ν™”Β·μ£Όλ³€ μ†ŒμŒ λ°˜μ‘μ΄ 30μ–΅ νŒŒλΌλ―Έν„° ν•˜λ‚˜μ— 톡합돼 μžˆμ–΄μš”.

μŒμ„± 벀치마크 MMAUμ—μ„œ 58.15점을 기둝해 기반 λͺ¨λΈμ„ λ„˜μ–΄μ„œκ³ , 7BκΈ‰ λŒ€ν˜• λͺ¨λΈμ—λ„ κ·Όμ ‘ν–ˆμ–΄μš”. 'μΈκ°„μ²˜λŸΌ λ“£κ³  κ°œμž…ν•˜λŠ” AI'κΉŒμ§€μ˜ 거리가 κ½€ μ’ν˜€μ‘Œλ„€μš”.

작돌쌀의 ν•œλ§ˆλ””

κΈ°μ‘΄ λͺ¨λΈμ€ λ…ΉμŒμ΄ λλ‚˜μ•Ό λ‹΅ν–ˆλŠ”λ°, 이제 μ‚¬λžŒμ²˜λŸΌ '흐름 μ†μ—μ„œ κ°œμž…'ν•˜λŠ” AIκ°€ μ˜€ν”ˆμ†ŒμŠ€λ‘œ ν’€λ Έμ–΄μš”.


좜처: New open-source voice model listens nonstop and decides every 0.4 seconds whether to speak or stay silent

이 글이 μ–΄λ• λ‚˜μš”?