Evaluating the intelligibility benefit of speech modifications in known noise conditions
Citations
107 citations
Cites background from "Evaluating the intelligibility bene..."
...9), did produce sparse representations which improved speech intelligibility by around 2 dB in a large-scale evaluation [60]....
[...]
73 citations
Cites methods or result from "Evaluating the intelligibility bene..."
...See [16] for more details....
[...]
...Note that this is not the same as the OptSII system described in [16]....
[...]
...Gains are expressed as equivalent intensity changes (EICs) computed by mapping scores to psychometric curves previously obtained for each masker using Plain speech (see [16] for details)....
[...]
...That study [16] compared 7 speech modification algorithms against read and Lombard speech and an unmodified TTS system....
[...]
59 citations
Cites background from "Evaluating the intelligibility bene..."
...8K utterances, respectively [77,78], all with fs = 16 kHz)....
[...]
47 citations
Cites background or methods from "Evaluating the intelligibility bene..."
...Moreover, this SS fixed filter has proven effective in an extensive evaluation of speech intelligibility enhancement modifications (Cooke et al., 2013)....
[...]
...Finally, an extensive evaluation of the ntelligibility of a variety of methods was recently carried out and described in Cooke et al. (2013)....
[...]
...In addition to the fixed filter, the SS described in Zorila et al. (2012) and evaluated in Cooke et al. (2013) also ncorporates adaptive components, including peak-sharpening (Hs(f)) and pre-emphasis (Hp(f)) filters....
[...]
47 citations
References
32,573 citations
"Evaluating the intelligibility bene..." refers background or methods in this paper
...The technique learns frequency band weights which maximise objective intelligibility using a genetic algorithm optimisation technique (Holland, 1975), with glimpse proportion (Cooke, 2006) as an objective intelligibility metric....
[...]
...Lombard sentences came from the same subset of the Harvard corpus as the plain material and were spoken by the same talker (see Section 3.1)....
[...]
8,837 citations
"Evaluating the intelligibility bene..." refers background or methods in this paper
...Lombard sentences came from the same subset of the Harvard corpus as the plain material and were spoken by the same talker (see Section 3.1)....
[...]
...After recording, all speech was downsampled from 96 to 16 kHz using Praat (Boersma, 2001), manually endpointed to remove leading and trailing silence and high-pass filtered with a cut-off frequency of 100 Hz to remove low-frequency artefacts....
[...]
2,169 citations
1,847 citations
"Evaluating the intelligibility bene..." refers methods in this paper
...Optimisation using other objective intelligibility or quality models (e.g. Christiansen et al., 2010; Taal et al., 2011; Rix et al., 2001) is likely to result in different modifications....
[...]
1,741 citations
"Evaluating the intelligibility bene..." refers methods in this paper
...The following parameters were used to train, adapt and generate speech: 59 Mel cepstral coefficients, Mel scale F0, and 25 aperiodicity energy bands extracted using STRAIGHT (Kawahara et al., 1999)....
[...]