Speech & Audio
1.[Paper Review] – LoopGen: Training-Free Loopable Music Generation
반복 재생되는 음악에서 전환의 부자연스러움을 해결하기 위해, circular padding과 beat alignment를 활용한 training-free 루프 생성 방식인 LoopGen을 제안
2.[Paper Review] – FlowSep: Fast and Accurate Language-Queried Sound Separation via Rectified Flow Matching
자연어로 원하는 소리만 분리해주는 FlowSep 모델 제안. Rectified Flow Matching 기반 생성 방식으로 빠르고 깔끔한 오디오 분리가 가능하다.
3.[Paper Review] Game-Oriented ASR Error Correction via RAG-Enhanced LLM
ASR for game domain
4.[Paper Review] VoXtream: Full-Stream Text-to-Speech with Extremely Low Latency
VoXtream / Streaming TTS / 102ms FPL
5.[Paper Review] moshi - temporal/depth transformer
moshi - temporal and depth transformer
6.[Paper Review] GLASS Flows
GLASS Flows: Transition Sampling for Alignment of Flow and Diffusion
7.[Paper Review] FSPEN: An Ultra-Lightweight Network for Real Time Speech Enhancement
출처: https://research.samsung.com/blog/FSPEN-AN-ULTRA-LIGHTWEIGHT-NETWORK-FOR-REAL-TIME-SPEECH-ENAHNCMENT 최근에 speech enhancement 분야를 보고 있는 중인데, 경량화된 모
8.DNS/URGENT 챌린지
DNS 챌린지와 URGENT 챌린지 소개
9.[Paper Review] – StreamFlow: Streaming Audio Generation from Discrete Tokens via Streaming Flow Matching
#streaming_decoder(≈vocoder)
10.[Paper Review] – Qwen3-TTS
qwen3-tts
11.kaldi 세팅하기
kaldi 세팅하기..
12.[Paper Review] Emotion Concepts and their Function in a Large Language Model
https://transformer-circuits.pub/2026/emotions/index.html 작성중..
13.화자 분리(Speaker Diarization) 기초 (1) - MFCC
시작하기 전에... 용어를 헷갈려 하실까봐 spectrum, spectrogram, mel spectrogram, mfcc 의 차이를 간단히 정리하자면 이와 같습니다. 이미지 출처 waveform → (pre-emphasis) → STFT(framing(hamming window, overlap, hop size) → 각 프레임에 DFT(실제로는 연산 ...
14.[Paper Review] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
https://arxiv.org/abs/2306.00978 작성중..
15.Full Duplex Model
정리중