iPlayer bit depth

[Bryn]

I was not sure whether to post this here or on the iPlayer thread, but decided hot to vitiate the latter with a somewhat esoteric matter.

I have long assumed that the HD Sound iPlayer steam for Radio 3 used a bit depth (quantization level) of 16 bits. However, if one captures the raw iPlayer stream AAL-LC offering to a file, then opens that file in a compatible DAW such as Sound Forge Pro 11, the quantization is presented as 32 bit (IEEE float). However, if I simply try to convert to WAV using "Free m4a to mp3 Converter 9.5" in its m4a to WAV setting, that results in file with 44.1kHz sample rate and 16 bit quantization. Since the original definitely had a sample rate of 48kHz, "Free m4a to mp3 Converter 9.5" clearly dong rather more conversion than it claims on the tin.

I am not well enough versed in the niceties of audio CODECs to have grasped that the AAC-LC specification includes the provision of variable quantization in addition to variable bit rate. So, the upshot is that if one wants to retain as high an audio quality as possible if converting such an AAC-LC file for burning to optical media, aim for 48kHz sample rate and 24 bit quantization (the greatest bit depth accepted by "Audio DVD Creator"). If burning to CD-R, it is probably best to downsample to 44.1kHz with anti-alias filtering and deduce the quantization to 16 bit using noise shaping.

[Cockney Sparrow]

Thanks Bryn - I think I am heading towards the area where your post will be relevant and helpful (in the absence of the wonderful lossless stream of last summer).

I'll decode the wording to understand it at the relevant time, I wondered in the past how the actual quality (beyond the BBC description) might be established. It might also help me to analyse the stream on the Naxos Music library via the Library. Although not wonderfiul (still a place for CD and FLAC downloads) it seems suspiciously better quality than its stated bit rate (although you may well be talking, above about something different from "bit rate" and I know you are not referring to mp3.

[Bryn]

Thanks Bryn - I think I am heading towards the area where your post will be relevant and helpful (in the absence of the wonderful lossless stream of last summer).

I'll decode the wording to understand it at the relevant time, I wondered in the past how the actual quality (beyond the BBC description) might be established. It might also help me to analyse the stream on the Naxos Music library via the Library. Although not wonderfiul (still a place for CD and FLAC downloads) it seems suspiciously better quality than its stated bit rate (although you may well be talking, above about something different from "bit rate" and I know you are not referring to mp3.

This issue of variable quantization (the number of bits used to represent the dynamic level of a sample) is new to me. Whether any of the 48 thousand sample per second in a given Radio AAC-LC stream is more than the 16 standard for CDs I do not know, but the possibility is why I am now opting to convert the 32 bit (IEEE floating) to 24 bit, rather than 16. It also saves me having to down-sample to 44.1kHz since the standard sample rate for DVD audio is 48kHz, the same as that of the Radio 3 AAC-LC stream.

By the way, the Audio DVD Creator application is pretty much legacy software. It is not directly compatible with versions of Windows after XP. However, I have found it possible to run it in "compatibility mode" and "as administrator". A bit of a faff, but it's such a useful program. At 48kHz sample rate and 24 bit quantization you get around 4 hours on a single DVD recordable.

[Dave2002]

Bryn - msg 1.

You raise several issues here. One fairly obvious question is "are there AAL-LC to xxxx converters which will do conversion (even a null conversion) to 48kHz and other sample rates - including 48 kHz?" I think that the AAL-LC software should be capable of coding and decoding to PCM or even DSD - and hence into a format which can be played or recorded on a DVD or CD. I think that converting to particular formats may well be something which individual software programs do, rather than something which is enshrined in an AAL-LC "standard". A comparison could be made with mp3, which is effectively capable of encoding more than 16 bit data depending on the encoding bit rates used. Arguably better results could be obtained from converting a higher bit depth output (say 20 bits depth) down to 16 bits (CD standard) by using appropriate dithering.

Although your assumption of a bit depth of 16 bits for the source for R3 HD may be correct - without insider information I have no way of knowing. It is perfectly possible that a lossy encoding scheme would effectively use more bits if it "thought" it worthwhile within the appropriate compression quality parameters, so the input sources might have more than 16 bits for bit depth.
Many such encoders use what are effectively sub-band coding, and may allocate a small number of bits - or even fractional bits (!!!) to the sub-bands.
Mathematically this works out, and the degree of compression is determined by the parameters and the coefficients in various matrices. Some compression is achieved by discarding small coefficients - though that's not the only compression method used. Another factor which affects some compression algorithms is whether the calculations are done using integer or floating point arithmetic.

I think you know all this already, but ....

[Bryn]

Bryn - msg 1.

You raise several issues here. One fairly obvious question is "are there AAL-LC to xxxx converters which will do conversion (even a null conversion) to 48kHz and other sample rates - including 48 kHz?" I think that the AAL-LC software should be capable of coding and decoding to PCM or even DSD - and hence into a format which can be played or recorded on a DVD or CD. I think that converting to particular formats may well be something which individual software programs do, rather than something which is enshrined in an AAL-LC "standard". A comparison could be made with mp3, which is effectively capable of encoding more than 16 bit data depending on the encoding bit rates used. Arguably better results could be obtained from converting a higher bit depth output (say 20 bits depth) down to 16 bits (CD standard) by using appropriate dithering.

Although your assumption of a bit depth of 16 bits for the source for R3 HD may be correct - without insider information I have no way of knowing. It is perfectly possible that a lossy encoding scheme would effectively use more bits if it "thought" it worthwhile within the appropriate compression quality parameters, so the input sources might have more than 16 bits for bit depth.
Many such encoders use what are effectively sub-band coding, and may allocate a small number of bits - or even fractional bits (!!!) to the sub-bands.
Mathematically this works out, and the degree of compression is determined by the parameters and the coefficients in various matrices. Some compression is achieved by discarding small coefficients - though that's not the only compression method used. Another factor which affects some compression algorithms is whether the calculations are done using integer or floating point arithmetic.

I think you know all this already, but ....

Until a couple of so years ago the iPlayer often used a sample rate of 44.1kHz for Radio 3, but now 48kHz is the standard. What was new to me was that the specification for AAC-LC includes a facility for processing an input signal at quantization levels varying to suit the particular sample. If I open an mp3 in Sound Forge, it presents as a fixed quantization level, usually 16 bit. It is aac files, whether raw or wrapped as m4a what open as 32 bit (IEEE floating). I think I will avoid using "Free m4a to mp3 converter" in future, now that I see it does rather more than simply resolving aac to WAV, but also resamples and bit rate converts. Who knows what artefacts it may introduce in the process?

[Dave2002]

I think I will avoid using "Free m4a to mp3 converter" in future, now that I see it does rather more than simply resolving aac to WAV, but also resamples and bit rate converts. Who knows what artefacts it may introduce in the process?

You could see (hear) what your ears tell you. I agree that using some "free" software is not always a good idea - particularly if you suspect problems. I have one - 5kplayer - which arguably is good for some video files, but I am very suspicious of it, and have disabled it from my machine.

If you can find a better converter, or at least one you're happy with then perhaps better to stick with that. Otherwise I suppose you could do some tests, and try to find out what artefacts the "Free m4a to mp3 converter" does generate so at least you'd know - though my guess is that there will be better tools out there.