What could be causing my canned lipsync to fail?

I'm struggling to get canned lipsync to work in Unity and would welcome any thoughts.

I have a character model, with animator, audio source, etc.

Its audio source has an OVRLipsyncContextMorphTarget, and the scene has an OVRLipSync helper object.

If I add an OVRLipSyncContext component, it works fine. So the morph target etc is set up ok.

The problem is that I need to use canned visemes for performance reasons.

So, I...

create a viseme file from an audio clip
replace the OVRLipSyncContext component with OVRLipSyncContextCanned, and assign the viseme file from 1 to its currentSequence field.

But when I play the audio clip, I get no lip movement at all.

A couple of things I've tried:

I've checked the viseme file: it's the right length and has viseme values which are large enough to produce visible movement.
I've added logging to OVRLipSyncContextCanned to verify that it is indeed running and copying the frame info in the Update loop (it is).

This all appears to be set up the same as the similar object in the Oculus LipSync demo scene.

Can anyone think of any gotchas that I might have missed?