Now that 5G is on the horizon, major urban areas will be racing to be the first to introduce their own smart city. AI in the telecom industry encompasses a myriad of focus areas. Customer service and network optimization are at the top of that list. Throughout this year, some of the biggest implementations of AI will be in these spaces this seems redundant. Predictive analytics is one of the most promising capabilities of AI technology, especially for the telecom community. AI-based analytics provide telecom companies with the ability to use data and machine learning software to predict future industry trends and potential infrastructure snags, and to continuously assess their service for bugs or potential pitfalls.
Additionally, AI-based chat-bots and virtual assistants will transform the way businesses handle customer service. They can automate conversations and successfully mimic human speech, while drawing data and insights to provide personalized and positive experiences.
Tech leader Mark Hurd believes that 85 percent of interactions with customers will be automated by Much like ecosystems function in nature, industrial ecosystems are a network of cross-industry organizations that work together to create a stronger, more advanced industry. The telecom industry will see a move toward creating those lucrative ecosystems.
Telecom companies have the highest potential to come together and form such alliances.
Do You Know the Average Mean Opinion Score in Your Network?
Industry leaders agree, with 83 percent citing ecosystems as an important piece of their disruption strategy. The introduction of IoT networks, backed with 5G capabilities, will make this even easier and will allow companies to seamlessly collaborate to advance these technologies across their networks.
The telecommunications industry is a vital center for cutting-edge technology adoption. With the integration of AI technologies, 5G networks, and the IoT, telecom organizations will continue to be at the forefront of technological growth. NetFortris provides flexible, scalable, cloud-based voice and data communications customized with specific security and compliance to give comprehensive solutions to business needs.
Multi-level protection, cloud-based backup, and bi-directional failover ensure business continuity while big data analytics with real-time service control improves business operational performance.
ViSQOL: an objective speech quality model
NetFortris also serves its customers with carrier-grade networks, business and communications applications, and expert service. The bits saved due to simplification of the pitch contour are used for the hidden transmission. Because of placing the approximation flag in the voice packet header, the improved version of the proposed algorithm yielded a significantly lower decrease in speech quality, when compared with the original version of HideF0.
In addition, for low bitrates of the hidden channel i.
Internet telephony during the last decade has become intensively used all over the world, and the VoIP traffic volume is constantly growing. Communication using IP telephony seems natural and constitutes a remarkable part of the data transmission on the Internet. It is no wonder that for a couple of years researchers have tried to use the VoIP traffic as a carrier for hidden transmissions.
Pitch, also known as the F 0 parameter, is one of the values that is extracted from the signal during speech encoding and is sent in the encoded bitstream as one of the parameters. Our novel steganographic method was based on approximating the pitch contour and using the saved bits to hide covert data. We also showed that this method was superior to a classical least significant bits LSB approach. This study is a continuation of research toward finding an efficient and secure steganographic method that can be used with Internet telephony.
In this article, we will present an improved version of the HideF0 algorithm, and we will test it with various narrowband modes of the Speex codec. We will also compare it with the previous version. Speech transmission over packet networks has been investigated as a carrier of hidden information for a couple of years, and so far, several approaches have been proposed. Methods based on packet header modification. Methods based on modification of packets' arrival time. Hybrid methods, which combine two or more of the aforementioned steganographic techniques.
Speech Quality Test and Verification
In the following subsections, these methods will be briefly characterized, and examples of their applications will be given. The methods based on voice payload modification directly control various bits within the data field of the voice packets. In order to do this without significant quality loss, these bits need to be carefully selected, and this is why these methods are usually tailored to a given speech codec. Numerous algorithms belonging to this group are based on the least significant bit LSB technique. In this approach, the sender of the steganogram takes control over the last i.
Properly selected LSBs can have either little or no impact on the output speech quality, and this is how they can be used to create a hidden data channel. One must remember, however, that such LSBs can also be easily identified and zeroed, to mitigate hidden transmission. Several steganographic methods were developed for the widely used G. Its main advantage is its simplicity — the codec does nothing but quantize each speech sample independently using a logarithmic quantizer. The authors researched which parameters could be optimally used for steganographic purposes based on LSB.
They proposed a solution that was able to create a hidden channel with a capacity of In , Huang et al. Yet another method of hiding data in the speech payload, called StegVoIP, was proposed by Tian et al. A codec. A different approach to voice payload modification was proposed by Mazurczyk et al. This method, called TranSteg Transcoding Steganography , consisted of replacing the original voice payload with the same speech data, but compressed at a higher compression ratio. The saved bits were used to carry a steganogram.
Huang et al. Algorithms belonging to this group affect time dependencies between voice packets, for example, by changing packet order. In , Chen et al.
The algorithm was based on the modification of the time distances between consecutive speech packets. A subset of algorithms that modify the voice packet payload exploit modifications of the fields that carry information about the F0 frequency. In this research, an LSB technique was applied to a pitch delay parameter of the G. The author took advantage of one of the psychoacoustic phenomena, such that the human ear is less sensitive to the same pitch difference in higher frequencies than in lower ones.
Using this knowledge, he elaborated a steganographic algorithm that would perform an adaptive quantization of the pitch parameter. Thanks to the saved bits, the algorithm was able to provide a steganographic bandwidth of ca.
P.862 and PESQ Output
In this way, at the expense of a maximum of 0. Its aim was to elaborate a speech codec that would work with high quality even in unreliable packet networks, such as the Internet. It is noteworthy that the G. The higher speech quality is offered, the larger dictionary of code vectors is used and a greater number of bits are required for IVQ.
The modes that offer good and very good speech quality, that is, modes between 3 and 7, use the so called fine pitch parameter FP and pitch gain PG , which are updated once per subframe. Speex offers additional features, such as transmission at a variable bit rate, that is, changing the bitrate dynamically, depending on the speech signal to be encoded e.
Another feature is discontinuous transmission in which the codec has the ability to switch to transmitting a comfort noise only or to stopping the transmission completely, to save the link usage and computational load. Besides its high speech quality, Speex is a free, open and easily available codec, so there should be no wonder that it is being more and more widely used. Contrary to this, in the current study, we propose a hybrid steganographic method that involves both header and payload modification. Such a change is motivated by the fact that the previous version contained an LSB component, which was both less secure and was suspected to have a negative impact on the output speech quality.
Using the proposed hybrid technique, in a way, we created two communication channels: one of the channels, using the header modification, has a signaling role — it will carry a flag that will inform if the second channel carries overt or hidden data. Generally, in our method, we wanted to take advantage of the fact that pitch is relatively slowly changing in its value for voiced speech, that is, when vocal folds are vibrating. When speech is unvoiced, that is, vocal folds are relaxed and are not vibrating, or when there is no speech activity at all during silence, breathing, etc.
We ran an experiment to verify the behavior of the FP parameter compared with F 0.
It can be seen that the FP takes random values in some frames, while it takes rather deterministic values in the others. When comparing the FP and the F 0 curves, one can notice that the deterministic segments of the FP do not always overlap with the steady segments of the F 0 curve. The mean square error E is measured between the actual FP values and the approximated ones. Then the approximation flag is set to zero, which in the current version of HideF0 is realized by proper modification of the packet header in the previous version it was obtained via an LSB of the last FP value in the frame.
First three original FP values within the frame can be removed and the hidden data can be embedded instead of them. The approximation flag in the VoIP packet header is set to one. In this study, we do not specify how to hide the approximation flag within an IP header or an RTP header — this can be realized in multiple ways.
If it equals zero, the frame is processed in the normal way. If the approximation flag equals one, the hidden data is extracted from the bits reserved for the three first FP values. The missing FP values are restored by approximating linearly between the last FP value of the preceding frame and the last FP value of the current frame.
It is important to mention that the HideF0 technique does not require the introduction of any additional delay, because the FP values are approximated using the data from the preceding frame. It means that the preceding FP value must be kept in the memory till the next iteration, which normally should not pose any problems.
Similar to the previous study, in the current experiments, the speech data used came from recordings extracted from TIMIT This corpus contains recordings originating from speakers from eight main dialects of American English. In our experiments, we used three male and three female speakers from each dialect, each uttering 10 phrases. In such a way, we created a data set of 24 male and 24 female speakers, with ca.
We decided to determine how HideF0 works with all the narrowband modes of the Speex codec, where HideF0 implementation is possible, that is, for modes starting from mode 3 called hereinafter Speex3 up to mode 7 called hereinafter Speex7. To estimate speech quality, PESQ required both the speech signal being tested and the original one. The steganographic cost, i. Output quality of the transcoded speech signal. Therefore, as previously, we used Speex in mode 5.
The distance between the two curves shown in this figure is impressive, especially in the region of low steganographic bandwidth — the newer version of HideF0 provides a much lower steganographic cost than the original algorithm.
Related Assessment and Prediction of Speech Quality in Telecommunications
Copyright 2019 - All Right Reserved