comparison doc/FR1-library-history @ 301:019eed8b1948

doc/FR1-library-history: new article
author Mychaela Falconia <falcon@freecalypso.org>
date Mon, 15 Apr 2024 18:36:21 +0000
parents doc/FR1-Rx-DTX@731c98b67da1
children 516e84085a15
comparison
equal deleted inserted replaced
300:be7ae099cbc3 301:019eed8b1948
1 The first Themyscira library for GSM-FR speech codec was libgsmfrp, an add-on
2 to legacy libgsm from TU-Berlin. The current offering is libgsmfr2, which
3 supplants the earlier combination of libgsm+libgsmfrp. This document details
4 the change history in this continuum of GSM-FR codec libraries.
5
6 Changes from libgsmfrp version 1.0.2 to libgsmfr2 version 2.0.0
7 ===============================================================
8
9 * Dependency on <gsm.h> defined types abolished, the entire library uses
10 <stdint.h> types instead.
11
12 * The Rx DTX handler component of the new library is unchanged from libgsmfrp
13 version 1.0.2, aside from the use of new types: uint8_t instead of gsm_byte,
14 explicit arrays of uint8_t instead of gsm_frame.
15
16 * In addition to this Rx DTX handler component, the new library includes the
17 GSM 06.10 encoder & decoder component (ported from libgsm pl22), the new
18 full decoder wrapper with decoder homing, an encoder homing function, and
19 new stateless frame packing and unpacking functions.
20
21 libgsmfrp change history: version 1.0.1 to version 1.0.2
22 ========================================================
23
24 There are only two changes, both involving corner cases with invalid SID frames
25 being received:
26
27 1) An invalid SID frame was received immediately following a good speech frame.
28 In this case we start CN generation, but we take the needed LARc and Xmaxc
29 parameters from the last speech frame, instead of the usual procedure of
30 extracting them from a valid SID frame. The change from 1.0.1 to 1.0.2
31 concerns the Xmaxc parameter in this corner case: in 1.0.1 we took Xmaxc
32 from the last subframe and used it for ensuing CN generation, but in 1.0.2
33 we compute a more proper mean Xmaxc from all 4 subframes, by dequantizing,
34 summing and requantizing.
35
36 2) An invalid SID frame was received in the speech muting state. The sequence
37 of inputs would have to be:
38
39 - a good speech frame;
40 - one or more BFIs, but not too many, so that the cached speech frame
41 does not decay fully by Xmaxc reduction;
42 - an invalid SID frame.
43
44 In version 1.0.1 we handled this even more obscure corner case by entering
45 the CN muting state, i.e., the state that is normally entered upon the
46 second lost SID. In version 1.0.2 we ignore invalid SID in the speech
47 muting state and act as if we got BFI, i.e., continue speech muting rather
48 than switch to CN muting.
49
50 libgsmfrp change history: version 1.0.0 to version 1.0.1
51 ========================================================
52
53 Version 1.0.0 exhibited the following defects, which are fixed in 1.0.1:
54
55 1) The last received valid SID was cached forever for the purpose of
56 handling future invalid SIDs - we could have received some valid
57 SID ages ago, then lots of speech or NO_DATA, and if we then get
58 an invalid SID, we would resurrect the last valid SID from ancient
59 history - a bad design. In our new design, we handle invalid SID
60 based on the current state, much like BFI.
61
62 2) GSM 06.11 spec says clearly that after the second lost SID
63 (received BFI=1 && TAF=1 in CN state) we need to gradually decrease
64 the output level, rather than jump directly to emitting silence
65 frames - we previously failed to implement such logic.
66
67 3) Per GSM 06.12 section 5.2, Xmaxc should be the same in all 4 subframes
68 in a SID frame. What should we do if we receive an otherwise valid
69 SID frame with different Xmaxc? Our previous approach would
70 replicate this Xmaxc oddity in every subsequent generated CN frame,
71 which is rather bad. In our new design, the very first CN frame
72 (which can be seen as a transformation of the SID frame itself)
73 retains the original 4 distinct Xmaxc, but all subsequent CN frames
74 are based on the Xmaxc from the last subframe of the most recent SID.