1 <?xml version="1.0" encoding="US-ASCII"?> 2 <!DOCTYPE rfc SYSTEM "rfc2629.dtd"> 3 <?rfc toc="yes"?> 4 <?rfc tocompact="yes"?> 5 <?rfc tocdepth="3"?> 6 <?rfc tocindent="yes"?> 7 <?rfc symrefs="yes"?> 8 <?rfc sortrefs="yes"?> 9 <?rfc comments="yes"?> 10 <?rfc inline="yes"?> 11 <?rfc compact="yes"?> 12 <?rfc subcompact="no"?> 13 <rfc category="std" docName="draft-ietf-codec-opus-update-02" 14 ipr="trust200902"> 15 <front> 16 <title abbrev="Opus Update">Updates to the Opus Audio Codec</title> 17 18 <author initials="JM" surname="Valin" fullname="Jean-Marc Valin"> 19 <organization>Mozilla Corporation</organization> 20 <address> 21 <postal> 22 <street>331 E. Evelyn Avenue</street> 23 <city>Mountain View</city> 24 <region>CA</region> 25 <code>94041</code> 26 <country>USA</country> 27 </postal> 28 <phone>+1 650 903-0800</phone> 29 <email>jmvalin (a] jmvalin.ca</email> 30 </address> 31 </author> 32 33 <author initials="K." surname="Vos" fullname="Koen Vos"> 34 <organization>vocTone</organization> 35 <address> 36 <postal> 37 <street></street> 38 <city></city> 39 <region></region> 40 <code></code> 41 <country></country> 42 </postal> 43 <phone></phone> 44 <email>koenvos74 (a] gmail.com</email> 45 </address> 46 </author> 47 48 49 50 <date day="1" month="July" year="2016" /> 51 52 <abstract> 53 <t>This document addresses minor issues that were found in the specification 54 of the Opus audio codec in <xref target="RFC6716">RFC 6716</xref>.</t> 55 </abstract> 56 </front> 57 58 <middle> 59 <section title="Introduction"> 60 <t>This document addresses minor issues that were discovered in the reference 61 implementation of the Opus codec that serves as the specification in 62 <xref target="RFC6716">RFC 6716</xref>. Only issues affecting the decoder are 63 listed here. An up-to-date implementation of the Opus encoder can be found at 64 http://opus-codec.org/. The updated specification remains fully compatible with 65 the original specification and only one of the changes results in any difference 66 in the audio output. 67 </t> 68 </section> 69 70 <section title="Terminology"> 71 <t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 72 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 73 document are to be interpreted as described in <xref 74 target="RFC2119">RFC 2119</xref>.</t> 75 </section> 76 77 <section title="Stereo State Reset in SILK"> 78 <t>The reference implementation does not reinitialize the stereo state 79 during a mode switch. The old stereo memory can produce a brief impulse 80 (i.e. single sample) in the decoded audio. This can be fixed by changing 81 silk/dec_API.c at line 72: 82 </t> 83 <figure> 84 <artwork><![CDATA[ 85 for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) { 86 ret = silk_init_decoder( &channel_state[ n ] ); 87 } 88 + silk_memset(&((silk_decoder *)decState)->sStereo, 0, 89 + sizeof(((silk_decoder *)decState)->sStereo)); 90 + /* Not strictly needed, but it's cleaner that way */ 91 + ((silk_decoder *)decState)->prev_decode_only_middle = 0; 92 93 return ret; 94 } 95 ]]></artwork> 96 </figure> 97 <t> 98 This change affects the normative part of the decoder, although the 99 amount of change is too small to make a significant impact on testvectors. 100 </t> 101 </section> 102 103 <section anchor="padding" title="Parsing of the Opus Packet Padding"> 104 <t>It was discovered that some invalid packets of very large size could trigger 105 an out-of-bounds read in the Opus packet parsing code responsible for padding. 106 This is due to an integer overflow if the signaled padding exceeds 2^31-1 bytes 107 (the actual packet may be smaller). The code can be fixed by applying the following 108 changes at line 596 of src/opus_decoder.c: 109 </t> 110 <figure> 111 <artwork><![CDATA[ 112 /* Padding flag is bit 6 */ 113 if (ch&0x40) 114 { 115 - int padding=0; 116 int p; 117 do { 118 if (len<=0) 119 return OPUS_INVALID_PACKET; 120 p = *data++; 121 len--; 122 - padding += p==255 ? 254: p; 123 + len -= p==255 ? 254: p; 124 } while (p==255); 125 - len -= padding; 126 } 127 ]]></artwork> 128 </figure> 129 <t>This packet parsing issue is limited to reading memory up 130 to about 60 kB beyond the compressed buffer. This can only be triggered 131 by a compressed packet more than about 16 MB long, so it's not a problem 132 for RTP. In theory, it <spanx style="emph">could</spanx> crash a file 133 decoder (e.g. Opus in Ogg) if the memory just after the incoming packet 134 is out-of-range, but our attempts to trigger such a crash in a production 135 application built using an affected version of the Opus decoder failed.</t> 136 </section> 137 138 <section anchor="resampler" title="Resampler buffer"> 139 <t>The SILK resampler had the following issues: 140 <list style="numbers"> 141 <t>The calls to memcpy() were using sizeof(opus_int32), but the type of the 142 local buffer was opus_int16.</t> 143 <t>Because the size was wrong, this potentially allowed the source 144 and destination regions of the memcpy() to overlap. 145 We <spanx style="emph">believe</spanx> that nSamplesIn is at least fs_in_khZ, 146 which is at least 8. 147 Since RESAMPLER_ORDER_FIR_12 is only 8, that should not be a problem once 148 the type size is fixed.</t> 149 <t>The size of the buffer used RESAMPLER_MAX_BATCH_SIZE_IN, but the 150 data stored in it was actually _twice_ the input batch size 151 (nSamplesIn<<1).</t> 152 </list></t> 153 <t> 154 The fact that the code never produced any error in testing (including when run under the 155 Valgrind memory debugger), suggests that in practice 156 the batch sizes are reasonable enough that none of the issues above 157 was ever a problem. However, proving that is non-obvious. 158 </t> 159 <t>The code can be fixed by applying the following changes to line 70 of silk/resampler_private_IIR_FIR.c: 160 </t> 161 <figure> 162 <artwork><![CDATA[ 163 ) 164 { 165 silk_resampler_state_struct *S = \ 166 (silk_resampler_state_struct *)SS; 167 opus_int32 nSamplesIn; 168 opus_int32 max_index_Q16, index_increment_Q16; 169 - opus_int16 buf[ RESAMPLER_MAX_BATCH_SIZE_IN + \ 170 RESAMPLER_ORDER_FIR_12 ]; 171 + opus_int16 buf[ 2*RESAMPLER_MAX_BATCH_SIZE_IN + \ 172 RESAMPLER_ORDER_FIR_12 ]; 173 174 /* Copy buffered samples to start of buffer */ 175 - silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 \ 176 * sizeof( opus_int32 ) ); 177 + silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 \ 178 * sizeof( opus_int16 ) ); 179 180 /* Iterate over blocks of frameSizeIn input samples */ 181 index_increment_Q16 = S->invRatio_Q16; 182 while( 1 ) { 183 nSamplesIn = silk_min( inLen, S->batchSize ); 184 185 /* Upsample 2x */ 186 silk_resampler_private_up2_HQ( S->sIIR, &buf[ \ 187 RESAMPLER_ORDER_FIR_12 ], in, nSamplesIn ); 188 189 max_index_Q16 = silk_LSHIFT32( nSamplesIn, 16 + 1 \ 190 ); /* + 1 because 2x upsampling */ 191 out = silk_resampler_private_IIR_FIR_INTERPOL( out, \ 192 buf, max_index_Q16, index_increment_Q16 ); 193 in += nSamplesIn; 194 inLen -= nSamplesIn; 195 196 if( inLen > 0 ) { 197 /* More iterations to do; copy last part of \ 198 filtered signal to beginning of buffer */ 199 - silk_memcpy( buf, &buf[ nSamplesIn << 1 ], \ 200 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) ); 201 + silk_memmove( buf, &buf[ nSamplesIn << 1 ], \ 202 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); 203 } else { 204 break; 205 } 206 } 207 208 /* Copy last part of filtered signal to the state for \ 209 the next call */ 210 - silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], \ 211 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) ); 212 + silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], \ 213 RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); 214 } 215 ]]></artwork> 216 </figure> 217 <t> 218 Note: due to RFC formatting conventions, lines exceeding the column width 219 in the patch above are split using a backslash character. The backslashes 220 at the end of a line and the white space at the beginning 221 of the following line are not part of the patch. A properly formatted patch 222 including the three changes above is available at 223 <eref target="https://jmvalin.ca/misc_stuff/opus_update.patch"/>. (EDITOR: 224 change to an ietf.org link when ready) 225 </t> 226 </section> 227 228 <section title="Downmix to Mono" anchor="stereo"> 229 <t>The last issue is not strictly a bug, but it is an issue that has been reported 230 when downmixing an Opus decoded stream to mono, whether this is done inside the decoder 231 or as a post-processing step on the stereo decoder output. Opus intensity stereo allows 232 optionally coding the two channels 180-degrees out of phase on a per-band basis. 233 This provides better stereo quality than forcing the two channels to be in phase, 234 but when the output is downmixed to mono, the energy in the affected bands is cancelled 235 sometimes resulting in audible artefacts. 236 </t> 237 <t>As a work-around for this issue, the decoder MAY choose not to apply the 180-degree 238 phase shift when the output is meant to be downmixed (inside or 239 outside of the decoder). 240 </t> 241 </section> 242 243 <section title="Hybrid Folding" anchor="folding"> 244 <t>When encoding in hybrid mode at low bitrate, we sometimes only have 245 enough bits to code a single CELT band (8 - 9.6 kHz). When that happens, 246 the second band (CELT band 18, from 9.6 to 12 kHz) cannot use folding 247 because it is wider than the amount already coded, and falls back to 248 LCG noise. Because it can also happen on transients (e.g. stops), it 249 can cause audible pre-echo. 250 </t> 251 <t> 252 To address the issue, we change the folding behaviour so that it is 253 never forced to fall back to LCG due to the first band not containing 254 enough coefficients to fold onto the second band. This 255 is achieved by simply repeating part of the first band in the folding 256 of the second band. This changes the code in celt/bands.c around line 1237: 257 </t> 258 <figure> 259 <artwork><![CDATA[ 260 b = 0; 261 } 262 263 - if (resynth && M*eBands[i]-N >= M*eBands[start] && \ 264 (update_lowband || lowband_offset==0)) 265 + if (resynth && (M*eBands[i]-N >= M*eBands[start] || \ 266 i==start+1) && (update_lowband || lowband_offset==0)) 267 lowband_offset = i; 268 269 + if (i == start+1) 270 + { 271 + int n1, n2; 272 + int offset; 273 + n1 = M*(eBands[start+1]-eBands[start]); 274 + n2 = M*(eBands[start+2]-eBands[start+1]); 275 + offset = M*eBands[start]; 276 + /* Duplicate enough of the first band folding data to \ 277 be able to fold the second band. 278 + Copies no data for CELT-only mode. */ 279 + OPUS_COPY(&norm[offset+n1], &norm[offset+2*n1 - n2], n2-n1); 280 + if (C==2) 281 + OPUS_COPY(&norm2[offset+n1], &norm2[offset+2*n1 - n2], \ 282 n2-n1); 283 + } 284 + 285 tf_change = tf_res[i]; 286 if (i>=m->effEBands) 287 { 288 ]]></artwork> 289 </figure> 290 291 <t> 292 as well as line 1260: 293 </t> 294 295 <figure> 296 <artwork><![CDATA[ 297 fold_start = lowband_offset; 298 while(M*eBands[--fold_start] > effective_lowband); 299 fold_end = lowband_offset-1; 300 - while(M*eBands[++fold_end] < effective_lowband+N); 301 + while(++fold_end < i && M*eBands[fold_end] < \ 302 effective_lowband+N); 303 x_cm = y_cm = 0; 304 fold_i = fold_start; do { 305 x_cm |= collapse_masks[fold_i*C+0]; 306 307 ]]></artwork> 308 </figure> 309 <t> 310 The fix does not impact compatibility, because the improvement does 311 not depend on the encoder doing anything special. There is also no 312 reasonable way for an encoder to use the original behaviour to 313 improve quality over the proposed change. 314 </t> 315 </section> 316 317 <section title="New Test Vectors"> 318 <t>Changes in <xref target="stereo"/> and <xref target="folding"/> have 319 sufficient impact on the testvectors to make them fail. For this reason, 320 this document also updates the Opus test vectors. The new test vectors now 321 include two decoded outputs for the same bitstream. The outputs with 322 suffix 'm' do not apply the CELT 180-degree phase shift as allowed in 323 <xref target="stereo"/>, while the outputs with suffix 's' do. An 324 implementation is compliant as long as it passes either the 'm' or the 325 's' set of vectors. 326 </t> 327 <t> 328 In addition, any Opus implementation 329 that passes the original test vectors from <xref target="RFC6716">RFC 6716</xref> 330 is still compliant with the Opus specification. However, newer implementations 331 SHOULD be based on the new test vectors rather than the old ones. 332 </t> 333 <t>The new test vectors are located at 334 <eref target="https://jmvalin.ca/misc_stuff/opus_newvectors.tar.gz"/>. (EDITOR: 335 change to an ietf.org link when ready) 336 </t> 337 </section> 338 339 <section anchor="IANA" title="IANA Considerations"> 340 <t>This document makes no request of IANA.</t> 341 342 <t>Note to RFC Editor: this section may be removed on publication as an 343 RFC.</t> 344 </section> 345 346 <section anchor="Acknowledgements" title="Acknowledgements"> 347 <t>We would like to thank Juri Aedla for reporting the issue with the parsing of 348 the Opus padding.</t> 349 </section> 350 </middle> 351 352 <back> 353 <references title="References"> 354 <?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml"?> 355 <?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.6716.xml"?> 356 357 358 </references> 359 </back> 360 </rfc> 361
