Showing appreciation, using this daily from the Read Aloud browser extension.

[kfatehi]

Hi @karim23657

Thanks for this great work you've done. Your talent, patience, and generosity is very much appreciated. I know this is not an easy project and it's really generous that you are publishing pretrained models. I thank you for that and want to let you know what I'm doing with it.

I am using one of your models to help me read Farsi websites online with the help of a chrome extension called Read Aloud. I made this little wrapper server https://github.com/kfatehi/persian-tts-server to run the model and access it via HTTP.

I'm still working out the proper design for people to be able to install the components easily for being able to access Coqui models such as this (in ken107/read-aloud#217 I'm trying to do some brainstorming), but I did want to just stop and let you know I'm a happy user of your pretrained model.

Specifically, I tried 3 of your pretrained models:

The first one, multispeaker vits, produced very bad results, but maybe I did something wrong. changiz, dilara, farid). I used code like this to generate 3 files, one for each speaker:

from TTS.config import load_config
from TTS.utils.manage import ModelManager
from TTS.utils.synthesizer import Synthesizer

model_path ="/kaggle/working/best_model_30824.pth"
config_path ="/kaggle/working/config.json"

# text="تاجر تو چه تجارت می کنی ، تو را چه که چه تجارت می کنم؟"
# text="​مراسم اهدا جایزه امین الضرب -1400به آقای حضرت حجت‌الاسلام والمسلمین سیدمحمود دعایی"
text="آرشیو روزانه"
synthesizer = Synthesizer(
    model_path, config_path
)

is_multi_speaker = False

if is_multi_speaker:
        
    # Get the list of available speaker names
    speaker_names = synthesizer.tts_model.speaker_manager.get_speakers()

    # Enumerate the speaker names
    for idx, speaker_name in enumerate(speaker_names):
        wavs = synthesizer.tts(text, speaker_name=speaker_name)
        synthesizer.save_wav(wavs, f'/outputs/{speaker_name}.wav')

else:
    wavs = synthesizer.tts(text)
    synthesizer.save_wav(wavs, f'/outputs/output.wav')

Next I downloaded the female vits and I think the voice sounds really great!

I am using it constantly now. I have difficulty reading new Farsi words, so a TTS is an absolute boon for a learner like myself to get acquianted with the sounds of words (and remember how to pronounce certain letters that I don't see often). Thank you thank you!!

Finally I tried male vits. It worked fine, but I didn't like the voice as much as the female one, so for now I am just using the female voice and I'm very happy with it.

I definitely look forward to more developments and higher quality more natural voices.

Thanks,
Keyvan