When duration is not reported in the files metadata, we directly infer the duration from the audio packets. This prevents errors on webm files. Very simple audio & video test files have been added that cover relevant usecases to prevent regressions.