Speaker Embeddings for Diarization of Broadcast Data In The Allies Challenge
| Autor: | Olivier Galibert, Anthony Larcher, Sylvain Meignier, Marie Tahon, Ambuj Mehrish, David Doukhan, Jean Carrive, Nicholas Evans |
|---|---|
| Přispěvatelé: | Laboratoire d'Informatique de l'Université du Mans (LIUM), Le Mans Université (UM), Institut National de l'Audiovisuel (INA), Laboratoire National de Métrologie et d'Essais [Trappes] (LNE ), Eurecom [Sophia Antipolis], ANR-19-CE23-0001,ExTENSoR,Réseaux de neurones évolutifs end-to-end pour la reconnaissance du locuteur(2019), ANR-17-CHR2-0004,ALLIES,Autonomous Lifelong learnIng intelLigent Systems(2017) |
| Rok vydání: | 2021 |
| Předmět: |
business.industry
Computer science Speech recognition Speech processing x-vectors raw signal [INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI] Speaker diarisation Speaker diarization Spectrogram Segmentation Mel-frequency cepstrum Sinc-Net business Cluster analysis Radio broadcasting Complement (set theory) i-vectors |
| Zdroj: | ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) ICASSP ICASSP, Jun 2021, Toronto, Canada |
| DOI: | 10.1109/icassp39728.2021.9414215 |
| Popis: | International audience; Diarization consists in the segmentation of speech signals and the clustering of homogeneous speaker segments. State-of-the-art systems typically operate upon speaker embeddings, such as ivectors or neural x-vectors, extracted from mel cepstral coefficients (MFCCs) or spectrograms. The recent SincNet architecture extracts x-vectors directly from raw speech signals. The work reported in this paper compares the performance of different embeddings extracted from MFCCs or the raw signal for speaker diarization and broadcast media treated with compression and sub-sampling, operations which typically degrade performance. Experiments are performed with the new ALLIES database that was designed to complement existing, publicly available French corpora of broadcast radio and TV shows. Results show that, in adverse conditions, with compression and sampling mismatch, SincNet x-vectors outperform i-vectors and x-vectors by relative DERs of 43% and 73% respectively. Additionally we found that SincNet x-vectors are not the absolute best embeddings but are more robust to data mismatch than others. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|