This work is part of research aimed to develop speech
technology for the Tatar language. First results for
pronunciation quality automatic assessing are presented.
Speakers are allowed for reading a predefined set of words
or sentences and the system tries to produce a reasonable
score. The pronunciation score for the entire utterance is
constructed counting for both the HMM-based log-
likelihood acoustic measure and the estimated duration of
the phoneme segment. The performance of the proposed
algorithms by measuring how well the machine-produced
scores correlate with human judgments are evaluated on a
speech corpus. Results and further research are discussed.