Tissue conductive unvoiced speech recognition in clean and noisy environments
| Autor: | Panikos Heracleous, Tomomi Kaino, Kiyohiro Shikano |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2005 |
| Popis: | In this paper, we present body-conductive unvoiced speech recognition in clean and noisy environments. Using a special microphone (e.g., NAM microphone) which is attached behind the talker's ear, we capture speech of very low intensity and inaudible to other listeners near the talker. The captured speech is called non-audible murmur (NAM) and it is unvoiced speech produced by vocal cords not vibrating. Moreover, body tissue and loss of lip radiation act as a low-pass filter and the high-frequency components are attenuated. However, the non-audible murmur spectral components still provide sufficient information to distinguish and recognize sounds accurately. A recognition system based non-audible murmur, can be used for privacy in human-computer communication. Moreover, due to the direct body transmission non-audible murmur recognition shows robustness against noise. To evaluate the performance of non-audible murmur recognition, we conducted experiments using clean and noisy data. Using a small amount of training data and adaptation approaches, we achieved in a clean environment a 93.9% word accuracy for a 20k vocabulary dictation task Using simulated noisy data, the performance did not decrease markedly. In real environments, however, the performance decreased leading to the Lombard reflex. SPECOM2005: 10th International Conference on Speech and Computer, October 17-19, 2005, Patras, Greece. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|