Best TTS you tried?

Posted by -Krag@reddit | LocalLLaMA | View on Reddit | 19 comments

I have been trying cocqui XTTS only. What is your experience?

[-]

WhisperBorderCollie@reddit

Best TTS was the RS+ version a few years ago. That thing was a whip

[-]

Tried Applio, made decent RVC models. For some reason the male voices turn out much better than female voices though. Biggest issue is it bounces back between British and American English and I don't know how to make it consistent.

[-]

ZeeRa2007@reddit

which one do you think is good for turning books to audiobook for personal use other than 11labs(nothing crazy)

[-]

redfairynotblue@reddit

Parler tts was trained on audiobooks. They have several voices you can choose from.

[-]

DataScientia@reddit

Recently some f5 tts was released and obviously eleven labs

[-]

Ok_Maize_3709@reddit

As someone building audio guides, I must say that nothing beats ElevenLabs in reliability. There are a lot of local TTS alternatives which are pretty cools but the issue is that sometime they get off the rails.

If not local, then I would say 1. ElevenLabs, 2. Azure TTS, 3. OpenAI (I mean the olde one, not the Advanced Voice Mode, it's a bit more robotic but better at emotion and intonations).

From the local ones F5TTS or XTTS2. Check out also Arena: https://huggingface.co/spaces/TTS-AGI/TTS-Arena

[-]

Charuru@reddit

Is there a ui to try f5, like a comfy node

[-]

lordpuddingcup@reddit

https://huggingface.co/spaces/mrfakename/E2-F5-TTS

[-]

chibop1@reddit

Yea it's pretty cool. Though I don't think it knows how to produce emotion from one reference audio. You still have to upload each reference audio for each emotion style. It just automates choosing the right reference audio based on the tag and feeding to the model to follow.

[-]