robustness of validation

[LoannPeurey]

Is your feature request related to a problem? Please describe.
Failure to read the sample rate of audios during the validation process results in a python error. This happens often on raw audio and should be more robust, as it forces to validate with --ignore-recordings

Describe the solution you'd like
solutions:

• implement new ways to read sample rate if the first one fails, or another better way
• continue validation even when not able to read sample rate (have a warning ??)
• a combination of both previous solutions?

[LoannPeurey]

PR #450 added a warning when can't read the sample rate, others ways of getting the sample rate may be explored before closing the issue