Encoding errors are not necessary critical

It could happen that encoding of a sentence fails, but there is no reason to assume that the same error happens with all sentences, so to abort the conversion when these errors are encountered is not always needed. Especially the difference with how text-to-speech failures are handled, is not logical. Text-to-speech is allowed to fail, resulting in missing audio (but no abortion of the conversion).

A better criterion for deciding to abort a conversion, is by looking at the ratio of sentences that have already gone wrong compared to the total.

Start by tracing where sentence-encoding failures abort conversion and where text-to-speech failures are tolerated. Compare the existing error handling and define tests for a sentence-failure ratio; done means isolated encoding errors do not always abort conversion, while the ratio-based abort behavior is covered.

Written by the indexing model from the issue text.