Peer-reviewed research
Joint Speech Recognition and Speaker Diarization via Sequence Transduction
Laurent El Shafey et al., Interspeech 2019, 2019
Read the originalRead on October 17, 2026
- What it is
- A Google research paper that trains one recurrent neural network transducer to do speech recognition and speaker diarization together, instead of running the two separately and merging the results. It reports results on a corpus of physician-patient conversations.
- How much weight it carries
- Peer-reviewed at Interspeech 2019. The test corpus is medical conversations recorded in one setting, and the baseline is the authors' own, so the size of the gain may not carry over to other audio.
Facts taken from this source
- In a 2019 Google study, a joint speech recognition and speaker diarization model improved word-level diarization error rate from 15.8 percent to 2.2 percent.
“our approach improves word-level diarization error rate from 15.8% to 2.2%” (Abstract)
- In the 2019 paper, a word counts toward word diarization error rate when it is a correctly recognized word with the wrong speaker label, or a substituted word with a wrong speaker label.
“CIS is the number of Correct ASR words with Incorrect Speaker tokens” (Section 5.2.5, equation 2)
- The 2019 paper says word diarization error rate must be used together with word error rate, because it does not account for deletions and insertions.
“Note that this WDER metric must be used in combination with the ASR Word Error Rate (WER) to account for deletions and insertions” (Section 5.2.5)