Добавил:
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:
Скачиваний:
0
Добавлен:
13.05.2026
Размер:
20.61 Mб
Скачать

4 | Discussion

In this chapter the different themes, RQs and results are discussed in order to paint a larger picture but also to highlight risks and possible shortcomings with the study. This is done in the upcoming five parts: Conclusion; Threats to validity; Ethics; Comparison to related work; and Future work.

4.1Conclusion

The following section will review every RQ and make some final conclusions in the light of the presented results, wrapping up with a summation at the end.

RQ1: How do different pairs of codecs and bitrates compare in terms of time for execution?

In the light of these results it seems evident that more compressed audio in general, and the LC3 library in particular, is faster when it comes to coding. By reducing the execution time it gives the developers more leeway when creating new solutions for BLE products to come. This does indeed look promising for the future implementations of the LC3 codec.

RQ2: How do compression with different pairs of codecs and bitrates compare in terms of perceived audio quality?

The bitrates in the audio test ranged from 64 to 2304 kbps, with the highest being the hidden reference. One differed in perceived audio quality compared to all others and that was LC3 (64), which is very low in this context of audio coders. If one would exclude 64 kbps it would not be a significant difference between any bitrate. When comparing the second lowest bitrate for LC3 (96 kbps) and the lowest for aptX (384 kbps) the amount of data for the former is merely 25% of the aptX solution.

The coders are doing a marvelous job and people have a hard time discerning the differences between these compressed and uncompressed audio samples. When it comes to LC3 in particular it seems to deliver a high perceived audio quality at very low bitrates.

34

RQ3: How does the attitude towards audio quality in a real-time interactive scenario compare to a non-interactive scenario?

Looking at people’s attitude towards audio quality, the audio demands are very similar for these scenarios, which indicates that the general idea of what is good in this context is quite unison. Hence, any manufacturer or developer of products that utilize this technology probably do not need to target a specific group of people. If it is good enough for one scenario it will probably cover several others as well.

Summary

When combining these three RQs it looks as though it is possible to come up with new codec implementations to compete in perceived audio quality with lower bitrates, whilst also satisfying the demands by multiple groups of people in different scenarios.

LC3 looks to be doing a good job paving the way for new possibilities moving forward and it will be interesting to see what these optimizations can be utilized for, whether it will be low latency audio transfers, increased battery life, solving new problems for society or just an increased QoS overall.

4.2Threats to validity

This part discusses the four types of validity threats, and their underlying nuances, that are presented by Wohlin et al. (2012) and how they might relate to this study.

Internal

The selection of participants for the experiment consisted of volunteers which can result in more motivated subjects. For the MUSHRA test this could mean that people had a tendency to find flaws in the audio samples when there were none.

When it comes to instrumentation the subjects might experience fatigue when listening for a long time which could have a slight effect in their results.

External

When considering the interaction of setting and treatment it cannot be excluded that the utilized hardware/firmware in the quasi-experiment favors a certain library/implementation. This is important to highlight since the goal is to generalize the idea that the different codecs differ in execution time regardless of the tools utilized to prove it.

There is also a risk that the sample size for the survey may not be large enough in order to generalize the results for all people who use audio devices in an interactive scenario.

It is a possibility that the audio sample used in the quasi-experiment favors one audio codec which could affect the generalizability of the results.

35

Conclusion

The reliability of measures is a possible threat since both the experiment and the survey was based on subjective measures. These are not as reliable as objective measures which can be repeated multiple times with the same outcome.

Construct

Another possible threat might have been inadequate preoperational explication of constructs, since the measured variable chosen for the survey was satisfaction level and this might not represent attitude entirely on its own.

There is also a risk for hypothesis guessing when it comes to the survey, since the shape of the questions might have given the subjects a hint of what was sought for, hence affecting their behavior/answers as a consequence.

In the experiment there was a risk of evaluation apprehension since some audio samples were nearly impossible to differentiate from one another. This in combination with being evaluated might have affected the participants’ behavior and thus spurred them to rate the samples lower or just differently, in order to look good instead of giving sincere and candid answers.

4.3Ethics

Two topics are mentioned under this section. The first relates to the participants in this research and how they were treated, referring to the definitions provided by Wohlin et al. (2012), whilst the second refers to the Sustainable Development Goals (SDGs) outlined by the United Nations (2015).

Participants

The confidentiality of the participants was considered early on. The only thing that can connect the subjects to the study are their full names and email addresses (which was needed for communication) but these have been kept hidden and are only known by the three authors conducting this research. I.e. no one else can know who participated and no other participant can know someone elses’ involvement unless they have communicated this with one another.

The results are disconnected from the subjects since this can be sensitive and a breach of privacy. Even if someone were to obtain the hidden raw data, which does not include any personal information, it would not be possible to link that information to any individual. When the results are presented there is no data indicating the sex, age or other personal information about the people involved since this was never asked for in the first place.

Upon taking part of the study a form of consent was approved and signed by every subject to ensure that they understood what was going on, and that they were comfortable with the handling of the data (Appendix C).

36

When inviting people to take part in the study a form of inducement was utilized in the form of fika plus the chance to win a cinema gift card (Appendix D). This was a simple way to make it more interesting for the participants as well as showing appreciation for their time and effort.

This study ultimately brings a scientific value by evaluating and spurring on the continuous development of wireless hearing devices. These are a big part of our modern society, utilized by people in all areas of life, and with an increased QoS this will most definitely open up new possibilities for how we all interact with our surroundings.

Sustainable development

Our research indicates that an optimized codec, like LC3, can ensure sustainable consumption and production patterns (SDG 12) by enabling products with low latency that are capable of satisfying users in multiple contexts. Thus potentially eliminating people’s need to have several headphones/earbuds for different scenarios.

By transferring audio wirelessly at lower bitrates it is possible to reduce energy consumption as this has a direct correlation with throughput (Giovanelli et al. 2015). Creating products that are more energy efficient will contribute in the effort to take urgent action to combat climate change and its impacts (SDG 13).

One main goal for LC3 is to include hearing aids in the Bluetooth domain (Schnell et al. 2021). As more smart and useful audio devices/solutions increase in popularity it benefits the hearing impaired as well, to ensure healthy lives and promote well-being for all at all ages (SDG 3).

If this technology can deliver decent audio quality at imperceivable latency it helps to reduce inequality within and among countries (SDG 10) by enabling more use cases, such as musicians playing together in real-time, without having professional and expensive gear related to In Ear Monitoring.

4.4Comparison to related work

This part will reflect on how our research relates to the previous mentioned articles under Related work (1.2).

The main difference between our research and the one from Hoene & Hyder (2010) is that they compare a different set of codecs, namely: SBC, CELT, aptX and ULD in an objective listening test called PEAQ. Their results showed that CELT outperformed all other codecs, including aptX, whilst our study concluded on a significant difference between LC3 at 64 kbps and all other pairs of codecs (LC3 and aptX) and bitrates. This could, to some extent, be explained by the different metrics produced by the PEAQ and the MUSHRA test, the first being an objective test and the latter a subjective, and thus no greater point should be made by this comparison.

In the second article Halbhuber et al. (2022) show that auditory delay has a negative effect on the players’ feelings towards the game. Hence, it highlights the importance of solving the issue of latency for hearing devices in real-time scenarios.

37

Соседние файлы в папке источники