previous | contents | next

Audio profiles and levels

In order to maximize interoperability, only a small number of profiles have been defined for MPEG-4 Audio. Given the rather large number of coding tools and object types, this leads to the inclusion of a relatively large number of audio object types even in the more simple profiles. Some of the audio profiles (e.g. main profile) contain both natural and structured audio object types. The following table lists only the audio profiles containing natural audio object types:
 

• Speech coding profile: This profile contains the CELP and HVXC object types as well as an interface to TTS (Text-To-Speech).
• Scalable profile: This profile contains the lower complexity AAC object types (LC, LTP), both in MPEG-2 (IS 13818-7) style and using the syntax which enables scalability (MPEG-4 style). In addition, the TwinVQ and speech coding object types and all the tools for scalable audio are part of this profile. It is expected, that most early applications will use this profile.
• Main profile: This is the do-it-all profile of MPEG-4 natural and structured audio. It contains all the MPEG-4 audio coding tools.

An hierarchical organisation of the profiles supports the design for interoperability: The speech coding profile is contained in all the other profiles containing natural audio coding tools, the scalable audio profile is contained in the main profile. Table V shoes all the tools of MPEG-4 natural audio and their use in the different audio objects types.
 

Table V: Usage of Audio Object Types

Audio Object Types	Tools									GA Bitstream Syntax Type	Hierachy
	13818-7 main	13818-7 LC	13818-7 SSR	PNS	LTP	TLSS	TwinVQ	CELP	HVXC
AAC main	X			X						ISO/IEC 13818-7 Style	contains AAC LC
AAC LC		X		X						ISO/IEC 13818-7 Style
AAC SSR			X	X						ISO/IEC 13818-7 Style
AAC LTP		X		X	X					ISO/IEC 13818-7 Style	contains AAC LC
AAC Scalable		X		X	X	X				scalable
TwinVQ					X		X			scalable
CELP								X
HVXC									X

 

Levels for the MPEG-4 audio scalable profile  

The large number of possibilities to combine different audio object types makes the traditional way of defining levels according to the channel count, sampling frequency etc. very difficult. In order to enable decoder implementers to conform with a certain level definition and still retain the possibility to combine different audio object types, complexity units have been defined and are used to calculate necessary decoder capabilities. For each audio object type, the decoder complexity (for a given sampling rate and channel count) was estimated in PCUs (computing complexity counted as millions of operations per second needed) and RCUs (memory complexity counted in kWords buffer requirements). Of course these complexity numbers depend a lot on the architecture of a decoder, whether realized on a dedicated DSP or a general purpose computing architecture. The following table lists the estimates of decoders for different object types as submitted to the MPEG audio group:
 
The level of a scalable profile decoder can now be determined by PCU and RCU numbers in addition to the number of channels and sampling frequencies. Four levels have been defined. They are:

• Level 1: One mono object of up to 24 kHz sampling frequency, all object types.
• Level 2: One stereo or two mono objects of up to 24 kHz sampling frequency.
• Level 3: One stereo or two mono objects of up to 48 kHz sampling frequency.
• Level 4: One 5.1 channel object or a exible configuration of objects up to 48 kHz sampling frequency and a PCU up to 30 and RCU up to 19.

Table VI: Decoder Complexity

Object Type	Parameters	PCU (MOPS)	RCU (kWords)
AAC Main 1)	fs=48 kHz	5	5
AAC LC 1)	fs=48 kHz	3	3
AAC SSR 1)	fs=48 kHz	4	3
AAC LTP 1)	fs=48 kHz	4	4
AAC Scalable 1) 2)	fs=48 kHz	5	4
TwinVQ 1)	fs=24 kHz	2	3
CELP	fs=8 kHz	1	1
CELP	fs=16 kHz	2	1
CELP	fs=8/16 kHz	3	1
HVXC	fs=8 kHz	2	1
Definitions: fs = sampling frequency   Notes: 1) PCU Proportional to sampling frequency 2) Includes core decoder

 

previous | contents | next