Strategies to Improve Robustness of Target Speech Extraction to Enrollment Variations

Satō, Hiroshi; Ochiai, Tsubasa; Delcroix, Marc; Kinoshita, Keisuke; Moriya, Takafumi; Makishima, Naoki; Ihori, Mana; Tanaka, Tomohiro; Masumura, Ryo

doi:10.21437/interspeech.2022-10767

Cited by 3 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The TSE module extracts the target-speaker's speech, X target , from mixture X using enrollment speech A clue . We use the time-domain TSE system we introduced in [12]- [14], which is similar to [30], [31].…”

Section: ) Tse Front-endmentioning

confidence: 99%

“…We used almost the same data as in previous works [12]- [14], and the difference is that the amount of data was increased. The details are shown in Table 1.…”

Section: A Experimental Setups 1) Datamentioning

confidence: 99%

“…To realize streaming TS-ASR, we need to 1) identify the target speaker in the mixture, 2) recognize that speaker while ignoring other speakers, and 3) realize this process in a streaming manner. We can realize TS-ASR by using a cascade of a target speech extraction (TSE) front-end with an ASR backend as in [12]- [14]. TSE focuses on extracting a single target speaker by conditioning the process on a pre-recorded enrollment utterance of the target speaker.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Streaming End-to-End Target-Speaker Automatic Speech Recognition and Activity Detection

et al. 2023

Self Cite

View full text Add to dashboard Cite

Automatic speech recognition of a target speaker in the presence of interfering speakers remains a challenging issue. One approach to tackle this problem is target-speaker speech recognition, which conditions the recognition process on an embedding that characterizes the voice of the target speaker. This enables recognizing only the speech of the target speaker while ignoring interferences. In this work, we propose an end-to-end target-speaker speech recognition system based on a neural transducer architecture to allow streaming and on-device recognition. Moreover, a target-speaker speech recognition system should be able to detect when the target speaker is inactive and output nothing in such a case. We introduce training and decoding schemes to allow target-speaker activity detection within our proposed recognition system. We confirm experimentally that our proposed end-to-end system performs competitively to conventional cascade approaches of a target speech extraction module and a recognition module while reducing computation costs and allowing streaming decoding.

show abstract

Section: ) Tse Front-endmentioning

confidence: 99%

“…We used almost the same data as in previous works [12]- [14], and the difference is that the amount of data was increased. The details are shown in Table 1.…”

Section: A Experimental Setups 1) Datamentioning

confidence: 99%