The present invention is concerned with the delivery, over a telecommunications link, of digitally coded material for presentation to a user.
In known systems of this type, a special server - often called a "streamer", controls delivery of material to a user terminal. Often, in the server, an item of material to be transmitted is stored as a single file; though
According to one aspect of the invention there is provided a terminal for playing audio or video material, as defined in claim 1.
In another aspect, the invention provides a method of transmitting digitally coded audio or video material, as defined in claim 14.
Other, optional, aspects of the invention are set out in the dependent claims.
Some embodiments of the present invention will now be described, with reference to the accompanying drawings, in which;
The system shown in
In these examples, the use of the hypertext transfer protocol is assumed; this is not essential, but is beneficial in allowing use of the authentication and security features (such as the Secure Sockets Layer) provided by that protocol.
Conventionally, a server for delivery of MP3 files takes the form of a so-called streamer which includes processing arrangements for the dynamic control of the rate at which data are transmitted depending on the replay requirements at the user terminal, for the masking of errors due to packet loss and, if user interaction is allowed, the control of the flow of data between server and client; here however the server 1 need contain no such provision. Thus it is merely an ordinary "web server".
The manner in which the data files are stored on the server 1 will now be explained. Suppose that an MP3-format file has been created and is to be stored on the server. Suppose that it is a recording of J. S. Bach's Toccata and Fugue in D minor (BWV565) which typically has a playing time of 9 minutes. Originally this would have been created as a single data file, and on a conventional streamer would be stored as this one single file. Here, however, the file is divided into smaller files before being stored on the server 1. We prefer that each of these smaller files is of a size corresponding to a fixed playing time, perhaps four seconds. With a compressed format such as MP3 this may mean that the files will be of different sizes in terms of the number of bits they actually contain. Thus the Bach file of 9 minutes duration would be divided into 135 smaller files each representing four seconds' playing time. In this example these are given file names which include a serial number indicative of their sequence in the original file, for example:
The partitioning of the file into these smaller sub-files may typically be performed by the person preparing the file for loading onto the web server 1. (The expression "sub-files" is used here to distinguish them from the original file containing the whole recording: it should however be emphasised that, as far as the server is concerned, each "sub-file" is just a file like any other file). The precise manner of their creation will be described more fully below. Once created, these sub-files are uploaded onto the server in a conventional manner just like any other file being loaded onto a web server. Of course the filename could also contain characters identifying the particular recording (the sub-file could also be "tagged" with additional information - when you play an MP3 file you get information on the author, copyright etc), but in this example the sub-files are stored on the server in a directory or folder specific to the particular recording - e.g. mp3_bwv565. Thus a sub-file, when required, may be requested in the form:
It is also convenient for the person preparing the sub-files for loading onto the server to create, for each recording, a link page (typically in html format) which is also stored on the server (perhaps with filename mp3_bwv565/link.htm), the structure and purpose of which will be described later.
It is also convenient that the web server stores one or more (html) menu pages (e.g. menu.htm) containing a list of recordings available, with hyperlinks to the corresponding link pages.
Turning now to the terminal, this may typically take the form of a conventional desktop computer, with, however, additional software for handling the reception of the audio files discussed. If desired, the terminal could take the form of a handheld computer, or even be incorporated into a mobile telephone. Thus
The sub-file naming convention used here, of a simple fixed length sequence of numbers starting with zero, is preferred as it is simple to implement, but any naming convention can be used provided the player program either contains (or is sent) the name of the first sub-file and an algorithm enabling it to calculate succeeding ones, or alternatively is sent a list of the filenames.
It will have been observed that the system described above offers the user no opportunity to intervene in the replay process. Nor does it offer any remedy for the possibility of buffer underflow (due for example to network congestion). Therefore a second, more sophisticated embodiment of the invention, now to be described, offers the following further features:
Note that these features are not interdependent, in that user control could be provided without rate-switching, or vice versa.
In order to provide for rate switching, the person preparing the file for loading onto the server prepares several source files - by encoding the same PCM file several times at different rates. He then partitions each source file into sub-files, as before. These can be loaded onto the server in separate directories corresponding to the different rate, as in the following example structure, where "008k", "024k" in the directory name indicates a rate of 8 kbit/s or 24 kbit/s and so on.
He also creates an index file (e.g. index.htm) the primary purpose of which is to provide a list of the data rates that are available.
index.htm
Note that because the length of a sub-file corresponds, as explained earlier, to a fixed length of time, the number of sub-files is the same for each directory. The subdirectory names comprise the data rate in kbit/s (three digits) plus the letter "k"; in this example indications of the audio sampling rate (11.025 kHz) and a mono-stereo flag are appended, for verification purposes.
The index file would thus contain a statement of the form :
(The <!--...--> simply indicates that the statement is embedded as a comment in an html file (or a simple text file could be used)). A typical index file is shown in
Initially the player program will begin by requesting, from the directory specified in the link file, the index file, and stores locally a list of available data rates for future reference. (It may explicitly request this file or just specify the directory: most servers default to index.htm if a filename is not specified.) It then begins to request the audio sub-files as described earlier, from the first-mentioned "rate" directory in the index file - viz. 024k_11_s (or the terminal could override this by modifying this to a default rate set locally for that terminal). The process from then on is that the player program measures the actual data rate being received from the server, averaged over a period of time (for example 30 seconds). It does this by timing every URL request; the transfer rate achieved (number of bits per second) between the client and server is determined. The accuracy of this figure improves as the number of requests goes up. The player maintains two stored parameters which indicate, respectively, the current rate, and the measured rate.
The initiation of a rate change is triggered:
The Buffer Low Percentage is the percentage of the time that the buffer contents represent less than 25% of the playout time (i.e. the buffer is getting close to being empty). If the Step Down Threshold is set to 0% then when the buffer empties the system always steps down when the other conditions are satisfied. Setting the Step Down Threshold to 5% (this is our preferred default value) means that if the buffer empties but the measured Buffer Low Percentage is greater than 5% it will not step down. Further buffer empties will obviously cause this measured rate to increase and will eventually empty the buffer again with a Buffer Low Percentage value exceeding 5% if the rate can not be sustained. Setting the value to 100% means the client will never step down.
The actual rate change is effected simply by the player program changing the relevant part of the sub-file address for example, changing "008k" to "024k" to increase the data rate from 8 to 24 kbit/s, and changing the current rate parameter to match. As a result, the next request to the server becomes a request for the higher (or lower) rate, and the sub-file from the new directory is received, decoded and entered into the buffer. The process just described is summarised in the following flowchart:
The user control is implemented by the user being offered on the screen the following options which he can select using the keyboard or other input device such as a mouse:
Alternatively, the user may be presented with a list of text labels or indices which may be selected (e.g. by means of a mouse) to initiate a jump. This might be implemented as follows:
The index.htm file held by the terminal, in memory, contains lines of the form (assuming that this information is embedded as comments within the document.
<-- !Odbits: Index = "0:01:44 Wireless" -->,
where where "Odbits:" is a keyword indicating to the player program that the following text is to be processed by a program, "Index" indicates the function that the player program is to perform, and the text with the quotation marks consists of the time (hours:minutes:seconds) from the beginning of the recording at which playing is to commence.
The player program reads the index.htm file and if it recognises an index command it generates a message ("an event") to the link.htm page which contains commands (typically written in Javascript) to handle such an event by displaying the labels at a desired place on the displayed page and respond to selection of those commands by generating a jump (containing the corresponding time) to the player program.
It is of interest to discuss further the process of partitioning the original file into sub-files. First, it should be noted that if (as in the first version described above), there is no expectation that a sub-file will be followed by a sub-file other than that which immediately follows it in the original sequence, then it matters little where the boundaries between the sub-files are located. In that case the sub-file size can be a fixed number of bits, or a fixed playing time length (or neither of these) - the only real decision is how big the sub-files should be. Where jumps are envisaged (in time, or between different data rates) there are other considerations. Where, as with many types of speech or audio coding (including MP3), the signal is coded in frames, a sub-file should contain a whole number of frames. In the case of rate switching, it is, if not actually essential, highly desirable that the sub-file boundaries are the same for each rate, so that the first sub-file received for a new rate continues from the same point in the recording that the last sub-file at the old rate ended. To arrange that every sub-file should represent the same fixed time period (e.g. the 4 seconds mentioned above) is not the only way of achieving this, but it is certainly the most convenient. Note however that, depending on the coding system in use, the requirement that a sub-file should contain a whole number of frames may mean that the playing duration of the sub-files does vary slightly. Note that in this embodiment of the invention, the available data rates, though they use different degrees of quantisation, and differ as to whether they encode in mono or stereo, all use the same audio sampling rate and in consequence the same frame size. Issues that need to be addressed when differing frame sizes are used are discussed below.
As for the actual sub-file length, excessively short sub-files should preferably be avoided because (a) they create extra network traffic in the form of more requests, and (b) on certain types of packet networks - including IP networks - they are wasteful in that they have to be conveyed by smaller packets so that overhead represented by the requesting process and the packet header is proportionately greater. On the other hand, excessively large sub-files are disadvantageous in requiring a larger buffer and in causing extra delay when starting play and/or when jumps or rate changes are invoked. A sub-file size of between 30% and 130% of the playout time, or preferably around half the playout time (as in the examples given above), is found to be satisfactory.
The actual process of converting the sub-files can be implemented by means of a computer programmed in accordance with the criteria discussed. Probably it will be convenient to do this on a separate computer, from which the sub-files can be uploaded to the server.
Another refinement that can be added is to substitute a more complex sub-file naming convention so as to increase security by making it more difficult for an unauthorised person to copy the sub-files and offer them on another server. One example is to generate the filenames using a pseudo-random sequence generator, e.g. producing filenames of the form :
In this case the player program would include an identical pseudo-random sequence generator. The server sends the first filename, or a "seed" of perhaps four digits, and the generator in the player can then synchronise its generator and generate the required sub-file names in the correct sequence.
In the above example of rate-switching, all the data rates used had the same frame size, specifically they used MP3 coding of PCM audio sampled at 11.025 KHz and a (PCM) frame size of 1152 samples. If it is desired to accomplish rate switching between MP3 (or other) recordings having different frame sizes, problems arise due to the requirement that a sub-file should contain a whole number of frames, because the frame boundaries do not then coincide. This problem can be solved by the following modified procedure for creating the sub-files. It should be noted particularly that this procedure can be used in any situation where rate switching is required and is not limited to the particular method of delivery discussed above.
Coding of the other data rates using different frame sizes proceeds in the same manner.
At the terminal, the control mechanisms are unchanged, but the decoding and buffering process is modified:
In this way, it is ensured that the sub-file sets for all rates have sub-file boundaries which correspond at the same points in the original PCM sample sequence.
Thus, each four-second period except the last is, prior to encoding, "padded" with audio samples from the next four-second period so as to bring the sub-file size up to a whole number of MP3 frames. If desired, the padding samples could be taken from the end of the preceding four-second period instead of (or as well as) the beginning of the following one.
Note that the MP3 standard allows (by a scheme known as "bit reservoir") certain information to be carried over from one audio frame to another. In the present context, while this is acceptable within a sub-file, it is not acceptable between sub-files. However, since naturally the standard does not allow such carry-over at the end or beginning of a recording, this problem is easily solved by encoding each sub-file separately, as if it were a single recording.
Changes of sampling rate (and indeed switching between mono and stereo operation) have some practical implications for operation of the audio interface 36. Many conventional sound cards, although capable of operation at a range of different settings, require re-setting in order to change sampling rate, and necessarily this causes an interruption in its audio output. Thus in a further modification, we propose that the sound card could be run continuously at the highest sampling rate envisaged. When the player program is supplying, to the buffer, data at a lower sampling rate, this data is then up-sampled to this highest rate before or after the buffer. Similarly, if the card is always operated in stereo mode, decoded mono signals can be fed in parallel so feed both the left and right channels of the sound card input. Again, if the number of bits per sample of the decoded signal is lower than expected by the card, the number of bits can be increased by padding with zeros.
Recollecting that the criteria discussed earlier for automatic data rate switching downwards envisaged a rate reduction only in cases of buffer underflow (involving therefore interruptions in the output), we note that with this modification such interruption can be avoided and therefore it is preferable to employ a criterion which anticipates underflow and avoids it in the majority of cases. In this case the first of the three AND conditions mentioned above (namely, that the buffer is empty) would be omitted.
Another feature that may be provided in the context of the system described is the display of visual information such as subtitles, or still pictures (in the nature of a slide show) to accompany the sound.
This might be implemented as follows:
The same principle may be applied to the delivery of video recordings, or of course, video recordings with an accompanying sound track. In the simpler version, where there is only one recording, the system differs from the audio version only in that the file is a video file (e.g. in H.261 or MPEG format) and the player program incorporates a video decoder. The manner of partitioning the file into sub-files is unchanged.
As in the audio case, there may be two or more recordings corresponding to different data rates, selected by the control mechanism already described. Also one can provide additional recordings corresponding to different replay modes such as fast forward or fast reverse which can be selected by an extension of the user control facilities already described. Again, a systematic convention for file and directory naming can be followed so that the player program can respond to - for example - a fast forward command by amending the sub-file address.
The delivery of video recordings does however have further implications for file partitioning if switching or jumps are to be permitted. In the case of recordings where each frame of a picture is coded independently, it is sufficient that a sub-file contains a whole number of frames of a picture. If compression involving inter-frame techniques is in use, however, the situation is more complex. Some such systems (for example the MPEG standards) generate a mixture of independently coded frames ("intra-frames") and predictively coded frames; in this case each sub-file should preferably begin with an intra-frame.
In the case of inter-frame coding systems such as the ITU H.261 standard, which do not provide for the frequent, regular inclusion of intra-frames, this is not possible. This is because - taking rate-switching as an example, if one were to request sub-file n of a higher bit rate recording followed by sub-file n+1 of a lower bit-rate recording, the first frame of the lower bit-rate sub-file would have been coded on an inter-frame basis using the last decoded frame of sub-file n of the lower rate recording, which of course the terminal does not have at its disposal - it has the last decoded frame of sub-file n of the higher rate recording. Thus serious mistracking of the decoder would occur.
In the case of switching between normal play and a fast play mode, the situation is in practice slightly different. On fast forward play at, for example, 5 times normal speed, one encodes only every 5th frame. In consequence the inter-frame correlation is much reduced and inter-frame coding becomes unattractive, so one would generally prefer to encode a fast play sequence as intra-frames. Switching from normal to fast then presents no problem, as the intra-frames can be decoded without difficulty. However, when reverting to normal play, the mistracking problem again occurs because the terminal is then presented with a predictively coded frame for which it does not have the preceding frame.
In either case the problem can be solved by using the principle described in our international patent application
The operation of this will now be described in the context of fast forward operation (fast rewind being similar but in reverse). In this example we assume that a 9 minute video sequence has been encoded at 96kbit/s according to the H.261 standard, and again at 5 times normal rate entirely at H.261 infra-frames, and that the resulting files have each been partitioned into four-second sub-files. Here, four seconds refers to the duration of the original video signal, not to the fast forward playing time. Following a naming convention similar to that employed above, the sub-files might be:
To switch from normal play to fast forward is only necessary for the player program to modify the sub-file address to point to the fast forward sequence - e.g.
Request mpg_ name/096k_x1/000055.bin
is followed by
Request mpg_name/096K_x5/000056.bin
In order to construct the bridging sequences for switching back to normal play it is necessary to construct a bridging sub-file for each possible- transition. As described in our international patent application mentioned above, a sequence of three or four frames is generally sufficient for bridging, so a simple method of implementation is to construct bridging sub-files of only 4 frames duration - e.g.
So that the switching is accomplished by a series of requests such as:
The bridging sub-file is generated as follows:
Exactly the same process could be used for rate-switching (albeit that now bridging sub-files are required in both directions). However, it will be observed that, as described, the bridging sub-file results in a freezing of the picture for a period of four frames - i.e. (at 25 frames per second) 160ms. In switching from fast to normal play this is acceptable - indeed one would probably choose to clear the buffer at this point. It may or may not be subjectively acceptable on rate-switching. An alternative therefore would be to construct a four-second bridging sequence. The request series would then look like:
The bridging sub-file would in that case be constructed either by recoding the fifth decoded frame of the decoded 128kbit/s sequence four times starting with decoded 96kbit/sframe 100,000 as the reference frame, or coding the first four frames of the decoded 128kbit/s sequence starting with decoded 96kbit/s frame 100,000 as the reference frame. In both cases the remaining 96 frames of the bridging sub-file would be a copy of the 128kbit/s sub-file.
The files to be delivered have been referred to as "recordings". However, it is not necessary that the entire audio or video sequence should have been encoded - or even exist - before delivery is commenced. Thus a computer could be provided to receive a live feed, to code it using the chosen coding scheme, and generate the sub-files "on the fly" and upload them to the server, so that, once a few sub-files are present on the server, delivery may commence.
One application of this delivery system would be for a voice-messaging system, as illustrated in
The same system can be used for a live audio (or video) feed. It is in a sense still "recorded" - the difference being primarily that delivery and replay commence before recording has finished, although naturally there is an inherent delay in that one must wait until at least one sub-file has been recorded and loaded onto the server 1.
The system can proceed as described above, and would be quite satisfactory except for the fact that replay would start at the beginning whereas what the user will most probably want is for it to start now - i.e. with the most recently created sub-file.
With a lengthy audio sequence one may choose to delete the older sub-files to save on storage: with a continuous feed (i.e. 24 hours a day) this will be inevitable and moreover one would need to reuse the sub-file names (in our prototype system we use 000000.bin to 009768.bin and then start again at 000000.bin), so that the older sub-files are constantly overwritten with the new ones. A simple method of ensuring delivery starting with the most recent sub-file would be to include in the index file an extra command instructing the player program to start by requesting the appropriate sub-file. This however has the disadvantage that the index file has to be modified very frequently -.ideally every time a new sub-file is created. Therefore we propose a method whereby the player program scans the server to find the starting sub-file, as follows. In the index file, the Mode parameter is set to "live" to trigger the player program to invoke this method. LFI is set to indicate the maximum number of sub-files that may be stored - say 9768. The method involves the following steps and presupposes that (as is conventional) each sub-file's "last modified" time and date has been determined. When using the HTTP protocol this can be achieved using a HEAD request which results not in delivery of the requested sub-file but only of header information indicating the time that the sub-file was written to the server, or zero if the sub-file does not exist. This time is represented below as GetURL(LiveIndex) where LiveIndex is the sequence number of the sub-file in question. Comments are preceded by "//".
1 LFI = 9768 // read from the index.htm file
LiveIndex = LFI / 2
StepSize = LFI / 2
LiveIndexModifiedAt = 0; // the beginning of time.
10 ThisIndexWasModifiedAt = GetURL(LiveIndex);
20 If (StepSize = 1)
{
// LiveIndexModifiedAt contains the time the file was written or 0 if no file
// has been found. LiveIndex contains the index.
goto 30
}
StepSize = StepSize / 2
if (ThisIndexWasModifiedAt > LiveIndexModifiedAt)
{
LiveIndexModifiedAt = ThisIndexWasModifiedAt;
LiveIndex = LiveIndex + StepSize
}
else
{
LiveIndex = LiveIndex - StepSize
}
Goto 10
30 FINISH
Having found the LiveIndex it is prudent to step back the Tp (playout time) and start to make the requests to fill the audio buffer from there. Playing may commence in the normal way.
Once the recording has actually finished, the index file can if desired be modified to set Mode to "recorded", and any length parameters.
If desired the player program could check periodically to see whether the index file has changed from "live" to "recorded" mode and if so to switch to "recorded" mode playing.
A simpler and much faster method of the identification of the "latest" sub-file will now be described, assuming, first of all, a single continuous sub-file numbering sequence.
This method is considerably faster than that described above for the cyclically numbered sub-files. Note that older sub-files may still be deleted, to reduce storage requirement, as long as the starting sub-file is kept. The method can however be modified to accommodate filename re-use (cyclic addresses), but would require:
There may be a danger of the playing operation running slightly faster or slower than the recording operation. To guard against the former it may be arranged that the player program checks each sub-file it receives to ascertain whether it is marked with a later time than the previous one: if not the sub-file is discarded and repeated requests made (perhaps three times) followed by a check of the index file if these requests are unsuccessful.
If the playing lags behind the recording process this can be identified by the player program occasionally checking the server for the existence of a significant number of sub-files more recent than those currently being requested, and if such sub-files do exist, initiating a "catching up" process - e.g. by regularly discarding a small amount of data.
This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.