The authors converted a speaker-recognition architecture, normally designed to represent complete utterances, into a causal frame-level encoder.
Frame-level speaker embeddings enable efficient CPU streaming diarization
FASTDIAR processes audio once, produces speaker embeddings every 80 milliseconds and clusters them online with a fixed 960-millisecond delay.
Big Tech
Nikita Torgashov · Okan Köpüklü
KTH Royal Institute of Technology · Microsoft
Research Digest··3 min read
Torgashov and Köpüklü present a streaming system for identifying who speaks when without repeatedly encoding overlapping audio windows.
Why this paper
From Microsoft and KTH Royal Institute of Technology · Released code
In one line
FASTDIAR streams speaker diarization on a CPU at five times real time with sub-second latency.
What it released
Code
What we could check
- ✓Code link in the paper (github.com)
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§