On Improving Telephone Speech Intelligibility for Hearing Impaired Persons Patrick Bauer 1 , Rosa-Linde Fischer 2 , Martina Bellanova 2 , Henning Puder 2 , Tim Fingscheidt 1 1 Institut für Nachrichtentechnik, Technische Universität Braunschweig, D-38106 Braunschweig 2 Siemens Audiologische Technik GmbH, D-91058 Erlangen 1 Email: {bauer,fingscheidt}@ifn.ing.tu-bs.de 2 Email: {rosa-linde.fischer,martina.bellanova,henning.puder}@siemens.com 1 Web: www.ifn.ing.tu-bs.de 2 Web: www.hearing.siemens.com Abstract Due to its limited acoustic bandwidth, conventional tele- phone speech is poorly intelligible, particularly for older and/or hearing impaired persons. Artificial bandwidth ex- tension (ABWE) aims at improving the intelligibility of narrowband (NB) speech by estimating and reconstructing the missing frequency components. Being employed at the receiver side, it does not require changes in the speech transmission system. Hence, it could be integrated into a telephone or directly into the hearing aid. This paper investigates the potential of ABWE to improve the intelli- gibility of NB telephone speech for hearing impaired per- sons wearing a hearing aid. Subjective listening tests on meaningless German logatomes demonstrate a significant improvement for critical fricatives, particularly for /s/. 1 Introduction The acoustic bandwidth of telephone speech is often still limited to a narrowband (NB) frequency range of about 300 ... 3400 Hz. Depending on the age, individual phys- iology and way of living, humans, however, are able to perceive frequencies between 20 Hz and 20 kHz. To al- low for wideband (WB) speech transmission in the range of 50 ... 7000 Hz, upcoming speech codecs are employed, such as the adaptive multirate wideband (AMR-WB) codec [1]. Until now, WB speech services are offered in Europe by only a few network providers 1 , because the required changes within the network infrastructure are quite expen- sive. In addition, a WB phone call will only be estab- lished, if the terminals on the receiver side and sending end are WB-capable. Since there are many NB-capable devices on the market yet, most phone calls would cur- rently fall back to a conventional NB mode, even if a WB telephone network was available. Hence, telephone speech will suffer from bandwidth limitations also in the foresee- able future. While the missing frequencies below 300 Hz are mainly responsible for a speech quality degradation [2], the frequency gap from 3.4 kHz to 7 kHz particularly re- duces speech intelligibility [3]. People without any hearing impairment may accept the limited bandwidth of telephone speech, but the older the people are, the more a bandwidth limitation will have an impact on their speech understanding capabilities [4, 5]. Even without any speech bandwidth limitation hearing im- paired persons of course suffer from intelligibility prob- lems. Therefore, they need a hearing aid that is individ- ually adapted to their specific hearing impairment. How- ever, hearing aids are so far not able to compensate for fur- 1 Deutsche Telekom AG announced on their homepage in November 2011 that the so-called HD Voice standard would be avail- able area-wide in T-Mobile’s 3G mobile networks. Network (2G/3G/4G, ISDN, VoIP) Sending End NB Speech Encoder A D Receiver Side D A ABWE NB Speech Decoder Figure 1: NB speech transmission system with artificial bandwidth extension (ABWE) located at the receiver side. ther bandwidth limitations that arise in a phone call. Sim- ply increasing the volume in order to elevate the speech level fails, because noise is thereby amplified as well and makes the bandlimited speech less intelligible [6]. For in- telligibility in noisy environments, it is particularly impor- tant for hearing impaired persons to enable WB speech transmission [7, Sec. 6.2.2]. In case of NB speech be- ing transmitted over the telephone channel, Fig. 1 shows, how missing frequency components can be estimated and reconstructed anyway by means of a technique called ar- tificial bandwidth extension (ABWE). It leaves the send- ing end as well as the telephone network completely un- touched. The only modification of the speech transmis- sion system is a modular ABWE block located at the re- ceiver side, e.g., in a telephone. To help hearing impaired persons feel more comfortable during phone calls, ABWE could also become part of their hearing aid. For this pur- pose, the telephone speech must be accessible. It may be received amongst others via the following wireless trans- mission paths [8]: 1. Acoustic transmission from the telephone loudspeaker to the microphone of the hearing aid, 2. Magnetic induction in the telecoil of the hearing aid caused by the voice coil of the telephone earphone, 3. Bluetooth streaming between the phone and a remote control with a relay station creating a near-field mag- netic induction in the telecoil of the hearing aids. In case of the first two options, hearing aids can only re- ceive the telephone speech monaurally, whereas the last option allows for a binaural reception. ABWE could be employed in all of these options. Since hearing aids gen- erally have installed individually adapted programs that shape the frequencies to compensate for the hearing impair- ment, it could directly be integrated as part of a telephone program (options 1–3). However, an outsourcing of ABWE into the remote control would be preferable in order to save the limited computing power of the hearing aid (option 3). This paper is structured as follows: Sec. 2 summarizes the proposed ABWE algorithm including offline training and real-time test process. In Sec. 3 logatome intelligibil- ity tests with hearing impaired subjects wearing a hearing aid are presented to demonstrate that ABWE is able to im- prove the intelligibility of NB telephone speech for critical fricatives. Conclusions are finally drawn in Sec. 4. ITG-Fachbericht 236: Sprachkommunikation · 26.-28.09.2012 in Braunschweig VDE VERLAG GMBH ∙ Berlin ∙ Offenbach 275