How can I use xtts for Turkish speech synthesis?

[trappedinspacetime]

I used to use the following line for Turkish tts:

tts --text 'on dokuz elli iki ' --model_name tts_models/tr/common-voice/glow-tts --out_path result.mp3

But the above model is poor quality for a Turkish tts.

I found the below new voice cloning xtts line which works very well:

    tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 --text "Bugün okula gitmek istemiyorum." --speaker_wav untitled.wav --language_idx tr

How can I use it for only text to speech in Turkish?

I tried something like :

 tts --text 'on dokuz elli iki' --model_name tts_models/multilingual/multi-dataset/xtts_v2 --out_path result.mp3 --speaker_idx tr

But it gives the following error:
> Using model: xtts

Text: on dokuz elli iki
Text splitted to sentences.
['on dokuz elli iki']
Traceback (most recent call last):
File "/home/kenn/.pyenv/versions/3.10.7/bin/tts", line 8, in
sys.exit(main())
File "/home/kenn/.pyenv/versions/3.10.7/lib/python3.10/site-packages/TTS/bin/synthesize.py", line 538, in main
wav = synthesizer.tts(
File "/home/kenn/.pyenv/versions/3.10.7/lib/python3.10/site-packages/TTS/utils/synthesizer.py", line 325, in tts
raise ValueError(
ValueError: [!] Missing speakers.json file path for selecting speaker tr.Define path for speaker.json if it is a multi-speaker model or remove defined speaker idx.

[erturkkadir]

@trappedinspacetime the following works fine here;

tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 --text "Bugün okula gitmek istemiyorum." --speaker_wav untitled.wav --language_idx tr

However, if I start server with

python3.10 TTS/server/server.py --use_cuda true --port 7865 --model_name tts_models/multilingual/multi-dataset/xtts_v2

then app crashed with following error;

File "/var/www/html/coqui/TTS/TTS/server/server.py", line 104, in
synthesizer = Synthesizer(
File "/var/www/html/coqui/TTS/TTS/utils/synthesizer.py", line 93, in init
self._load_tts(tts_checkpoint, tts_config_path, use_cuda)
File "/var/www/html/coqui/TTS/TTS/utils/synthesizer.py", line 183, in _load_tts
self.tts_config = load_config(tts_config_path)
File "/var/www/html/coqui/TTS/TTS/config/init.py", line 82, in load_config
ext = os.path.splitext(config_path)[1]
File "/usr/lib/python3.10/posixpath.py", line 118, in splitext
p = os.fspath(p)
TypeError: expected str, bytes or os.PathLike object, not NoneType

one more thing, I was able to get response from other model (tts_models/tr/common-voice/glow-tts) with following cURL
curl -X POST XXX.XXX.XXX/.XXX:XXX/api/tts --header "language-id: tur" --header "style-wav: referans.wav" --header "speaker-id: tur" -d "text=okula gitmek istemiyorum&language-id=tur" --output snd.wav

Hope you found a good solution

[trappedinspacetime]

Unfortunately, no. I hope somebody finds a solution.

[Rumeysakeskin]

@trappedinspacetime , If you want to use tts for a specific language,

1. You can check all Coqui available speakers with the following command:

!pip install TTS
!tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \
    --list_speaker_idx

2. You can do inference using one of the available speakers using the following command:

!tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \
     --text "Bugün okula gitmek istemiyorum." \
     --speaker_idx "Asya Anara" \
     --language_idx tr \
     --use_cuda true

import IPython
IPython.display.Audio("/content/tts_output.wav")