Tongue movement detection cannot track only six phonemes: - b - p - m - f - v - h

That’s a remarkably small number compared with other methods. English could easily be optimized to achieve a sub-5% WER with a very small model.