Output of --list_speaker_idxs for XTTS should not include latents

[eginhard]

When I try to find speakers list by tts --list_speaker_idxs --model_name tts_models/multilingual/multi-dataset/xtts_v2, it just send me list of embeddings, but not speakers list.

[ 1.4573e-02]]])}, 'Damjan Chapman': {'gpt_cond_latent': tensor([[[-0.1542, 1.0372, 0.0600, ..., -0.4648, -0.6890, 2.2647],
[-0.3850, 0.1696, 1.0869, ..., -0.3312, 0.7753, 1.5265],
[-0.4134, 0.6549, -1.0348, ..., -0.2458, -0.6853, 1.4418],
...,
[-0.4208, 0.5191, -0.6756, ..., -0.2182, -0.5243, 1.1794],
[-0.6379, 1.5274, 0.2272, ..., -1.5168, -0.2494, 1.1695],
[ 0.3209, -0.0421, 0.7555, ..., 0.8700, -0.3047, 2.2061]]]), 'speaker_embedding': tensor([[[-8.7840e-03],
[ 1.7883e-02],

Originally posted by @pivolan in #138

We should make the output more useful in this case, e.g. just listing the speaker names.