Source for an IETF Internet-Draft defining mouth-to-ear response latency (MRL) for conversational voice systems, together with an active method for measuring it at the RTP egress reference point of the calling endpoint.
MRL is the interval between the transmission of the final speech sample of a caller's utterance and the arrival of the first sample of the system's response audio. Two variants are defined, one taken at packet arrival and one taken behind a de-jitter buffer of stated target depth, and both are required in any report. The method is specified so that both timestamps come from a single clock on a single host, which means the metric needs no synchronisation between the measuring endpoint and the system under test.
Pre-submission. This draft has not been submitted to the IETF, has no working group behind it, and carries unresolved editorial questions listed in its final appendix. The most significant of those is what should count as response audio when a system emits a filled pause or an earcon before its substantive reply.
The intended venue is the IPPM working group, though the charter question is open and DISPATCH and the Independent Submission stream are both plausible alternatives.
The toolchain is deliberately kept outside any project virtual environment:
python3 -m venv ~/.venvs/ietf-tools
~/.venvs/ietf-tools/bin/pip install xml2rfc
gem install kramdown-rfc
Then:
make # renders draft-nygate-ippm-mrl-00.txt
make html # renders the HTML version
make check # house style checks
make clean
The KRAMDOWN path at the top of the Makefile assumes Homebrew Ruby and will need
adjusting if the Ruby version changes.
An open instrument implementing this method is published separately at dnygate/voice-ai-latency-harness, archived at 10.5281/zenodo.22124823. Where the implementation and this specification disagree, the specification is correct and the implementation has a bug.
Contributions to the IETF are made under the terms of BCP 78 and BCP 79.