Towards Singing Voice Enhancement from Larynx Microphone Signals (en)

* Presenting author
Day / Time: 08.03.2023, 08:40-09:00
Room: Saal Y8
Typ: Vortrag (strukturierte Sitzung)
Abstract: Larynx microphones provide a practical way to obtain interference-free recordings of the human voice by picking up vibrations directly from the throat with a piezo-electric sensor. While being particularly suitable for use cases like radio communication in noisy environments, larynx microphones have also demonstrated their value in musicological research, e.g., for analysing individual voices in polyphonic singing. However, the recorded signals do not provide satisfactory audio quality for mixing and playback, as the effects of the vocal tract (like vowel formants or consonants) are barely present in the recorded signal and the frequency response is limited. In this contribution, we introduce an approach for singing voice enhancement from larynx microphone signals using methods from differentiable digital signal processing (DDSP). In particular, we train a neural network to control a filter and synthesis model that enhances the larynx microphone input signal to sound more like a traditional close-up microphone recording. Additionally, we provide a suitable dataset of musical performances with approx. 210 minutes of audio recordings from five individual singers, where both larynx and close-up microphone signals are available. Finally, we evaluate the subjective quality of our approach with a listening test and discuss possible paths towards further improvements.


strpos(): Passing null to parameter #1 ($haystack) of type string is deprecated

/kunden/431424_13355/webseiten/subdomains/, line 35
32.              */
33.             if (headers_sent()) {
34.                 $encoding = false;
35.             } elseif (strpos($HTTP_ACCEPT_ENCODING, 'x-gzip') !== false) {
36.                 $encoding = 'x-gzip';
37.             } elseif (strpos($HTTP_ACCEPT_ENCODING, 'gzip') !== false) {
38.                 $encoding = 'gzip';