OriginIndian systems are built by companies based in India. International systems are built by companies
based outside India.
Comparative scoreA standard statistical model for head-to-head contests, called Bradley-Terry, turns
thousands of evaluator wins and losses into one strength score per system. It belongs to the same family of
methods used for chess ratings and chatbot arenas. Beating a strong system counts for more than beating a weak
one. A score of 0 is the average across the systems tested, so a negative score means below that average.
± rangeEvery score comes with a 95% confidence range: the span within which the system's true score is
very likely to fall. We estimated it by re-running the analysis 1,000 times, each time on a different random
draw of evaluators. Systems are ranked in order of their score, but where two ranges overlap, the gap between
those systems is not statistically clear. Further tests, corrected for the number of comparisons made, show
where the real gaps lie.
Mean Opinion ScoreA Mean Opinion Score (MOS) measures how good audio sounds to people. An evaluator hears
one clip on its own and rates it on a scale of 1 to 5, and the score is the average of those ratings. In
MicCheck, evaluators rate each clip on six dimensions of audio quality.
OrdinaryEveryday sentences that fall outside the other four types, selected so that their sounds occur as
often as they do in the language (phonetically balanced).
Code-mixedSentences that mix in words from another language, as people naturally do in everyday speech.
NumeralsSentences containing digits in any script, or patterns such as phone numbers, dates, ordinals
like "3rd", ranges, prices, percentages and decimals.
QuestionsSentences that end in a question mark (? or ؟), where the voice must carry question intonation.
Long sentencesThe longest sentences in each language, where pacing and pausing matter most.