Try this:

SearchClient.Instance.Conventions.UnifiedSearchRegistry.ForInstanceOf<MediaData>()
.ProjectExcerptUsing<ISearchContent>(spec =>
doc =>
!string.IsNullOrWhiteSpace(doc.SearchAttachmentText)
? doc.SearchAttachmentText.AsCropped(spec.ExcerptLength)
: doc.SearchAttachment.AsCropped(spec.ExcerptLength)
);

By your description,

Test - SearchAttachment$$attachment` (base64)

Acceptation - `SearchAttachmentText$$string` <-- This is what you want. 

I would say that it is your test env that is having problem parsing and indexing pdf contents. 

Thank you for your response. Your answer helped me to the solution and it was very close. We're using HitSpecification with HighlightExcerpt property set to true, so that means that we needed another method: one for the highlighting. Below is the solution I came up with.

    public class FindInitialization : IInitializableModule
    {
        public void Initialize(InitializationEngine context)
        {
            SearchClient.Instance.Conventions.UnifiedSearchRegistry.ForInstanceOf<MediaData>()
                .ProjectHighlightedExcerptUsing(HighlightExpressionGetter);
        }

        private static Expression<Func<ISearchContent, string>> HighlightExpressionGetter(HitSpecification spec)
        {
            var highlightSpec = new HighlightSpec
            {
                FragmentSize = spec.ExcerptLength,
                NumberOfFragments = 1,
                PreTag = spec.PreTagForAllHighlights,
                PostTag = spec.PostTagForAllHighlights
            };

            return doc => !string.IsNullOrWhiteSpace(doc.SearchAttachmentText) ?
                    doc.SearchAttachmentText.AsHighlighted(highlightSpec) : doc.SearchAttachment.AsHighlighted(highlightSpec);
        }
    }