RTP Payload Formats for European Telecommunications Standards Institute (ETSI) European Standard ES 202 050, ES 202 211, and ES 202 212 Distributed Speech Recognition Encoding
Extracted elements (23)
DSR speech recognizers are extremely sensitive to consecutive frame losses; therefore one SHOULD limit the maximum number of DSR frames per packet or employ FEC or interleaving when high packet loss is expected. This is why maxptime defaults to 80ms but lower values are explicitly recommended for lossy paths.
ES 202 211 and ES 202 212 extensions transmit additional pitch (Pidx) and voicing class (Cidx) information alongside mel-cepstral features to enable speech waveform reconstruction for human audition and to improve recognition of tonal languages such as Mandarin, Cantonese, and Thai.
The feature frame size of DSR presented to RTP packetization is not dependent on the number of speech samples in each 10 ms frame; rather, it is fixed by the codec's output vector format. This decouples RTP packetization from the underlying sampling rate.
All SDP parameters in this payload format are declarative and all reasonable values are expected to be supported; the standard Offer/Answer model per RFC 3264 applies without special handling.
ES 202 050 uses a slightly different frame format from ES 201 108 and the two do not interoperate with each other. The ES 202 050 RTP payload format uses the same RTP packet layout as RFC 3557 (ES 201 108), but differs in DSR codec frame bit definition and payload type MIME registration.
Any number of FPs MAY be aggregated together in an RTP payload and they MUST be consecutive in time. One SHOULD always keep the RTP payload size smaller than the MTU to avoid IP fragmentation.
Congestion control for RTP MUST be used in accordance with RFC 3550 and any applicable RTP profile such as RFC 3551.
For ES 202 050, pairs of quantized 10 ms mel-cepstral frames MUST be grouped together and protected with a 4-bit CRC forming a 92-bit FP, then padded with 4 zeros to the MSB 4 bits of the last octet to achieve octet alignment.
For ES 202 211, the mel-cepstral frame format MUST be used as defined in Section 6.2.4 of ES 202 211, and each FP MUST be padded with 4 zeros to the MSB 4 bits of the last octet for octet alignment.
For ES 202 212, the 4-bit CRC and 2-bit PC-CRC in the FP MUST be calculated using the formula (including bit-order rules) defined in Section 7.2.4 of ES 202 212, and each FP MUST be padded with 4 zeros to the MSB 4 bits of the last octet.
The 4-bit CRC and 2-bit PC-CRC in the ES 202 211 FP MUST be calculated using the formula (including bit-order rules) defined in Section 6.2.4 of ES 202 211.
The 4-bit CRC in the ES 202 050 FP MUST be calculated using the formula (including bit-order rules) defined in Section 7.2 of ES 202 050.
The RTP header marker bit (M) MUST be set following the general rules for audio codecs, as defined in Section 4.1 of RFC 3551.
A Null FP for ES 202 050 is defined by setting the content of both frames to null (filling the first 88 bits with zeros), computing the 4-bit CRC per Section 7.2.4 of ES 202 050, and padding 4 zeros to achieve octet alignment. Null FPs mark the end of a transmission segment.
A Null FP for ES 202 211 is defined by setting all 112 bits of the FP to zero, and is sent to mark the end of a transmission segment.
A Null FP for ES 202 212 is defined by setting all 112 bits of the FP to zero, sent to mark the end of a transmission segment.
SDP mapping: MIME type 'audio' maps to the SDP 'm=' media name; MIME subtype maps to 'a=rtpmap' encoding name; 'rate' maps to the clock rate in 'a=rtpmap'; 'ptime' and 'maxptime' map to 'a=ptime' and 'a=maxptime' attributes respectively.
The RTP timestamp corresponds to the sampling instant of the first sample encoded for the first FP in the packet. The timestamp clock frequency equals the sampling frequency (in samples). The timestamp increases by 160, 220, or 320 for each consecutive FP at 8, 11, or 16 kHz respectively.
Three new MIME audio subtypes were registered with IANA: 'dsr-es202050' for ES 202 050 AFE, 'dsr-es202211' for ES 202 211 XFE, and 'dsr-es202212' for ES 202 212 XAFE. All have optional parameters: 'rate' (8000, 11000, or 16000; default 8000), 'maxptime' (default 80ms), and 'ptime'.
Implementations using these payload formats are subject to the security considerations of RFC 3550 (RTP) and applicable RTP profiles such as RFC 3551. No additional security services beyond those are specified.
The ES 202 050 Frame Pair (FP) consists of two 44-bit mel-cepstral frames plus a 4-bit CRC, padded with 4 zero bits to the MSB of the last octet, totaling 96 bits (12 octets) per FP. Each FP represents 20 ms of speech.
The ES 202 211 FP consists of two 44-bit codebook index frames, a 4-bit CRC, 7-bit Pidx1, 5-bit Pidx2, 1-bit Cidx1, 1-bit Cidx2, 2-bit PC-CRC, and 4-bit padding — totaling 112 bits (14 octets) per FP representing 20 ms of speech.
The ES 202 212 FP has the same bit layout as ES 202 211: two 44-bit codebook index frames with VAD flag in each, 4-bit CRC, 7-bit Pidx1, 5-bit Pidx2, 1-bit Cidx1, 1-bit Cidx2, 2-bit PC-CRC, and 4-bit padding — totaling 112 bits (14 octets) per FP.