Blog about voice enhancements

Always a wideband experience – Part 2

Always a wideband experience – Part 2

In a series of two articles, VoiceGurus are looking into HD voice and the benefits of Bandwidth Extension technology. Previously, we looked at the market today. This is part two, where we will discuss how Bandwidth Extension technology enhances the quality of narrowband calls to bring back that wideband experience.

1. Bandwidth Extension always provides a wideband experience

To prevent consumer dissatisfaction, Bandwidth Extension is a solution that can be integrated in mobile phones by device makers. In case the phone receives a narrowband signal from the network, it will automatically generate a wideband signal. It re-creates the high frequency part of the spectrum between 3.7 and 8 kHz that was removed by the network. Figure 1 shows how Bandwidth Extension can be integrated in the Rx – downlink – audio path of a mobile phone.

Part 2 - Figure 1 - Integration Architecture PPTX

Figure 1: Integration of Bandwidth Extension into mobile phone audio path

Looking at the spectral content of the human voice up to 4 kHz, you can predict reliably what the higher frequency spectrum above 4 kHz must have looked like. This technique is called spectral folding and when done accurately, it reconstructs the speech signal very closely to the original. In most Bandwidth Extension solutions there is a trade-off between the amount of high-frequency spectral content being added and the amount of artifacts being introduced. These artifacts are caused by errors in the spectral folding process and are mostly audible in fricative sounds such as ‘s’, ‘v’ and ‘f’. If the Bandwidth Extension algorithm is tuned too aggressively, the artifacts start to overshadow the benefits of the wideband sound, making the overall result even worse than the narrowband original. The challenge of Bandwidth Extension is to create powerful, crisp high-band speech between 4 and 8 kHz and at the same time protect the fricative sounds to prevent annoying artifacts.

2. ACR quality measurements

The quality of a Bandwidth Extension solution can be measured using Absolute Category Rating (ACR) tests. When carried out under controlled laboratory conditions, ACR tests allow accurate unbiased comparisons of Bandwidth Extension solutions against narrowband and wideband reference signals. Using Mean Objective Scoring (MOS) each contender in the benchmark receives a score on a scale from 1 to 5. Describing such an ACR test would be worth a VoiceGurus article by itself, but the short summary would be that a state-of-the-art Bandwidth Extension would approximate real HD Voice quality by at least 80%.

3. Other quality parameters of Bandwidth Extension solutions

Speech quality is the most important element of performance of a Bandwidth Extension solution, but there are other aspects worth to consider. As mentioned before, narrowband phone calls do not have a bandwidth of 4 kHz but slightly lower. Low-pass filters in sample rate converters and speech codecs like AMR-NB cannot reach the full 4 kHz bandwidth and start dropping their frequency response above 3.7 kHz. For humans, this is not very noticeable when calling, but Bandwidth Extension solutions attempting to fold the spectrum around the 4 kHz mark might create a gap in the output spectrum. Smart Bandwidth Extension solutions will take special precautions to also restore the gap to avoid unnecessary coloring of the output speech signal.
Another aspect of Bandwidth Extension is the sensitivity to background noise in the Rx input signal. The Bandwidth Extension process is usually preceded by a noise suppression algorithm, to clean up the signal and to prevent background noise from being folded into the high band. However, any noise not removed by the noise suppressor should be replicated correctly, as if it were part of a natural speech signal.
Related to this is the dependency on a certain speech codec. Speech in mobile communications undergoes a number of encoding-decoding cycles when travelling from one caller to the other, inevitably degrading the speech quality by inserting noise or by making the voice sound ‘metallic’. Ideally, Bandwidth Extension solutions should make no assumptions on the speech codec being used by the network because the speech codecs used by the network are not always known.
Furthermore, Bandwidth Extension solutions should be language independent. They should support both tonal languages like Mandarin Chinese or Thai as well as non-tonal languages such as most European languages.

4. What do mobile operators say about Bandwidth Extension?

Some operators are already mandating the use of Bandwidth Extension in some cases. For example, AT&T in the US mandates the use of Bandwidth Extension on all VoLTE-capable devices in their latest Device Requirements specification. Other operators are following or are considering adding device requirements for Bandwidth Extension. Anyhow, the objective is always the same: To provide a continuous wideband experience to consumers, during a phone call and between different calls.

5. Commercial example: VoiceExperience from NXP Software

One example of Bandwidth Extension can be found in LifeVibes™ VoiceExperience 5.0 from NXP Software. It makes every phone call sound like a wideband call. If it’s an HD Voice call you get of course HD Voice quality. Narrowband calls will be enhanced to sound like if it were an HD Voice call. The audio clip below gives an example of what that sounds like.

Part 2 - Figure 2 - Bandwidth Extension Intro (NS ON) 16 kHz EQ 6 dB graphs PPTX

Figure 2: Audio example comparing Bandwidth Extension against narrowband and HD Voice

Figure 3 – Bandwidth Extension Intro (NS ON) 16 kHz EQ 6 dB audio clip.mp3

6. Conclusion

Bandwidth Extension is a valuable solution to overcome today’s limitations of HD Voice. It provides a continuous wideband experience between two phone calls and even during a call. It is currently being tested by device makers and mobile network operators and it is expected to hit the market in the second half of 2014. Stay tuned: Once commercial devices are commercially available, VoiceGurus plan to perform a comparative benchmark like we did for Noise Suppression recently.

Leave a Reply