Is this AlignTTS the best it can be with my custom dataset?

[zubairahmed-ai]

I have an 1h 34mins dataset that I am training AlignTTS with, the voice is still sounding a bit robotic, but loss isn't dropping anymore as you see below. Please let me know if this is due to a smaller dataset or some config change that I can try. I have previously turned off noam_schedule but loss would gradually increase (light blue line below)

My config.json and test audios

Now the stats with noam_schedule turned off (light blue line)