about pocket-tts

[Danik-droid]

I use pocket-tts with vulkan and 7900xtx in cachyos. Overall it runs extremely fast, but I'm having trouble configuring more than one voice.

These are the voices I would like to use.

I use this to run program:
./build/linux-vulkan-release/bin/audiocpp_server --config server.json --backend vulkan

my config file:
server.json

{ "host": "127.0.0.1", "port": 8090, "backend": "vulkan", "device": 1, "threads": 15, "lazy_load": true, "models": [ { "id": "pocket-tts", "family": "pocket_tts", "path": "/run/media/daniel/Data/LLM/models/TTS/pocket-tts/", "task": "tts", "mode": "offline", "load_options": { "language": "english" }, "session_options": { "language": "english" }, "default_voice_preset": { "voice_id": "azelma" }, "voice_presets": { "cosette": { "voice_id": "cosette" } }, "voice_presets": { "anna": { "voice_id": "anna" } }, "voice_presets": { "alba": { "voice_id": "alba" } } } ] }

my SilyTaver settings:

Problem: only one voice works, the one I choose first—always. What am I doing wrong?

[0xShug0]

@Danik-droid Sorry for the late reply! I didn't get a notification for the discussion post... I'll look into it.

[0xShug0]

@Danik-droid It's the config issue. The right config is one "voice_presets" key and multiple voices.

{
"default_voice_preset": {
"voice_id": "azelma"
},
"voice_presets": {
"cosette": {
"voice_id": "cosette"
},
"anna": {
"voice_id": "anna"
},
"alba": {
"voice_id": "alba"
}
}
}

[Danik-droid]

Like that?

{ "host": "127.0.0.1", "port": 8090, "backend": "vulkan", "device": 1, "threads": 15, "lazy_load": true, "models": [ { "id": "pocket-tts", "family": "pocket_tts", "path": "/run/media/daniel/Data/LLM/models/TTS/pocket-tts/", "task": "tts", "mode": "offline", "load_options": { "language": "english" }, "session_options": { "language": "english" }, "default_voice_preset": { "voice_id": "azelma" }, "voice_presets": { "cosette": { "voice_id": "cosette" }, "anna": { "voice_id": "anna" }, "alba": { "voice_id": "alba" } } } ] }

Still work only one voice.

[0xShug0]

@Danik-droid Would you mind sharing a few more details about your test setup? In your tests, are you sending requests directly to the server, or are you using a third-party UI?

I sent the requests to the server directly (output attached):

curl -fsS http://127.0.0.1:8090/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -o azelma.wav \
  -d '{"model":"pocket-tts","input":"Testing the Azelma voice preset.","voice":"azelma","seed":1234}'

curl -fsS http://127.0.0.1:8090/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -o cosette.wav \
  -d '{"model":"pocket-tts","input":"Testing the Cosette voice preset.","voice":"cosette","seed":1234}'

curl -fsS http://127.0.0.1:8090/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -o anna.wav \
  -d '{"model":"pocket-tts","input":"Testing the Anna voice preset.","voice":"anna","seed":1234}'

curl -fsS http://127.0.0.1:8090/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -o alba.wav \
  -d '{"model":"pocket-tts","input":"Testing the Alba voice preset.","voice":"alba","seed":1234}'

curl -fsS http://127.0.0.1:8090/v1/audio/speech \
  -H 'Content-Type: application/json' \
  -o default_voice.wav \
  -d '{"model":"pocket-tts","input":"Testing the default voice preset.","seed":1234}'

alba.wav
anna.wav
azelma.wav
cosette.wav
default_voice.wav

[Danik-droid]

I'm using https://github.com/SillyTavern/SillyTavern
with that settings:
https://private-user-images.githubusercontent.com/179402314/626802903-3f21e252-72e6-493f-8490-34c26715c3b7.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3ODUyNTkyNjIsIm5iZiI6MTc4NTI1ODk2MiwicGF0aCI6Ii8xNzk0MDIzMTQvNjI2ODAyOTAzLTNmMjFlMjUyLTcyZTYtNDkzZi04NDkwLTM0YzI2NzE1YzNiNy5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYwNzI4JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MDcyOFQxNzE2MDJaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT0xMmIwNGY1MGY5YzJhNDI5NWZhYzk1N2QyMzVlYWE1NjhkY2FlYWYyZmQ4MjIzMDdjNDI5ZWUzYTBhZDgwZjUwJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZyZXNwb25zZS1jb250ZW50LXR5cGU9aW1hZ2UlMkZwbmcifQ.YmBXF_fyHtbiSr5ByNKdJ8WFQkHSNmc4YculnMYP-mQ

In fact, it also works for me when I send directly.

So maybe it's a bad SillyTavern configuration... I don't know.

[0xShug0]

Thanks for confirming! Looks like a SillyTavern issue. Please consider submitting an issue there