annotate doc/AMR-EFR-philosophy @ 336:7f99b8ed30e5

libtwamr: integrate gc_pred.c
author Mychaela Falconia <falcon@freecalypso.org>
date Thu, 18 Apr 2024 23:02:35 +0000
parents 83408f67a96c
children 9bcf65088006
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
311
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
1 Relation between GSM-EFR and 12k2 mode of AMR
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
2 =============================================
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
3
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
4 What are the differences between GSM-EFR codec and the highest 12k2 mode of AMR,
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
5 or MR122 for short? The most obvious difference is in DTX: the format of SID
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
6 frames and even the very paradigm of how DTX works are completely different
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
7 between EFR and AMR. But what about non-DTX operation? If a codec session
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
8 consists solely of good speech frames, no SIDs and no BFI frame gaps, are EFR
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
9 and MR122 strictly identical?
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
10
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
11 The correct answer is that in the absence of SIDs, EFR and MR122 are directly
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
12 interoperable in that the output of an EFR encoder can be fed to the input of
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
13 an AMR decoder, and vice-versa. However, the two codecs are NOT identical at
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
14 the bit-exact level! The differences are subtle, such that finding them
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
15 requires some intense study; this article documents some of these study
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
16 findings:
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
17
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
18 https://www.freecalypso.org/hg/efr-experiments/file/tip/Theory-and-mystery
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
19
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
20 What other DSP/transcoder vendors have done
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
21 ===========================================
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
22
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
23 ETSI had a tradition of defining standard GSM codecs (FR, HR, EFR) in bit-exact
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
24 form, and every production implementation was required to match the output of
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
25 the official reference bit for bit. However, once AMR came out, the regulation
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
26 on EFR was loosened. GSM 06.54 document from 2000-08 (ETSI TS 100 725 V5.2.0)
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
27 has an appendix-like chapter (chapter 10) whose first paragraph reads:
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
28
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
29 The 12.2 kbit/s mode of the Adaptive Multi Rate speech coder described
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
30 in TS 26.071 is functionally equivalent to the GSM Enhanced Full Rate
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
31 speech coder. An alternative implementation of the Enhanced Full Rate
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
32 speech service based on the 12.2 kbit/s mode of the Adaptive Multi Rate
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
33 coder is allowed. Alternative implementations shall implement the
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
34 functionality specified in TS 26.071 for the 12.2 kbit/s mode, with the
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
35 exception that the DTX transmission format (GSM 06.81) and the comfort
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
36 noise generation (GSM 06.62) shall be used.
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
37
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
38 It appears that DSP vendors (for GSM MS or for network transcoders, or perhaps
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
39 both) weren't too happy with the prospect of having to include two different
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
40 versions of _almost_ the same codec algorithm with a bunch of interspersed
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
41 subtle diffs, and so the rules were bent: EFR implementors were given permission
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
42 to deviate from the original bit-exact definition of EFR in order to have more
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
43 commonality with MR122.
136
8eb0e7a39409 doc: document command line utilities
Mychaela Falconia <falcon@freecalypso.org>
parents:
diff changeset
44
311
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
45 Approach adopted for Themyscira GSM codec libraries suite
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
46 =========================================================
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
47
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
48 I (Mother Mychaela) previously entertained the idea of creating a unified codec
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
49 library that supports both AMR and EFR with common code, producing a published-
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
50 source, FOSS-culture equivalent of what most proprietary vendors have done.
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
51 However, on further reflection, that idea has been rejected. The current vision
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
52 (as of 2024-04) is that libgsmefr (stable since early 2023) and libtwamr
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
53 (currently a work in progress) shall remain separate and independent libraries,
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
54 the former implementing GSM-EFR (the original bit-exact definition) and the
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
55 latter implementing AMR. My reasons for this decision are:
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
56
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
57 * Libgsmefr already exists, and it is already a bit of a jewel compared to the
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
58 sorry state of true GSM codec support in the world of FOSS outside Themyscira.
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
59 Giving up on this library and moving to some nebulous new one does not sound
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
60 appealing.
136
8eb0e7a39409 doc: document command line utilities
Mychaela Falconia <falcon@freecalypso.org>
parents:
diff changeset
61
311
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
62 * There does not exist any formal, bit-exact definition for what we informally
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
63 call "EFR version 2": the realization of EFR as implemented by post-AMR-era
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
64 proprietary vendors, some sort of AMR-EFR hybrid. As I see it, it is not my
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
65 place to try to innovate in speech codec design, instead it is my job to
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
66 provide 100% correct, bit-exact implementations of existing solid standards -
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
67 and there is no bit-exact standard to follow for "EFR version 2".
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
68
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
69 * Libtwamr project: the task of turning the original AMR code from 3GPP into a
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
70 proper library, style-consistent with Themyscira libgsmfr2 and libgsmefr,
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
71 without the ugliness of opencore-amr, is already a lot of work as it is.
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
72 There is no need to make it harder by adding the task of supporting AMR-based
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
73 EFR, especially when the latter lacks formal definition.
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
74
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
75 Performance issues
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
76 ==================
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
77
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
78 Right now the only significant downside of libgsmefr compared to
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
79 libopencore-amrnb is that our library is significantly slower: almost 7 times
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
80 slower on non-DTX encode and a little over 3 times slower on SID-free decode.
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
81 However, this performance problem will need to be solved by profiling the code
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
82 to find the slowest spots, comparing the code of individual blocks between ours
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
83 and theirs, and porting over whatever performance-optimizing strategies were
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
84 implemented in OpenCORE code base. The latter code base is a derivative work
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
85 based on 3GPP AMR source, hence the guts of the codec are largely the same
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
86 between 3GPP AMR and libopencore-amrnb; the latter has been significantly
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
87 performance-optimized, but also heavily uglified. But there is no reason why
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
88 the same performance fixes can't be applied to EFR code base - it will simply
83408f67a96c doc/AMR-EFR-philosophy: new article
Mychaela Falconia <falcon@freecalypso.org>
parents: 136
diff changeset
89 take work. This work is currently part of our future roadmap.