Question about chatterbox.tts_turbo and cfg_weights and exageration

[credo99]

Hi, in the presets.yaml of the ui there are several configurations that are displayed in the turbo model page, all of them have among the parameters both cfg_weights and exageration but in the server log there is this warning: "chatterbox.tts_turbo: CFG, min_p and exaggeration are not supported by Turbo version and will be ignored". My questions are 1) Are these two parameters ignored in the case of turbo model usage? 2) Arethese configuration valid only for the original chatterbox model usage?

Thanks

[appsteve]

Hi, in the presets.yaml of the ui there are several configurations that are displayed in the turbo model page, all of them have among the parameters both cfg_weights and exageration but in the server log there is this warning: "chatterbox.tts_turbo: CFG, min_p and exaggeration are not supported by Turbo version and will be ignored". My questions are 1) Are these two parameters ignored in the case of turbo model usage? 2) Arethese configuration valid only for the original chatterbox model usage?

Thanks

Thanks for bringing this up! Yeah, it looks like the Turbo version is not showing the CFG and Exaggeration settings, so not sure if it's using them or defaulting them to certain values. The TTS quality of the Turbo version also seems to be less gooder (LOL), or lower quality, that the original version. There needs to be more clarity around this.

[credo99]

Hi, in the presets.yaml of the ui there are several configurations that are displayed in the turbo model page, all of them have among the parameters both cfg_weights and exageration but in the server log there is this warning: "chatterbox.tts_turbo: CFG, min_p and exaggeration are not supported by Turbo version and will be ignored". My questions are 1) Are these two parameters ignored in the case of turbo model usage? 2) Arethese configuration valid only for the original chatterbox model usage?
Thanks

Thanks for bringing this up! Yeah, it looks like the Turbo version is not showing the CFG and Exaggeration settings, so not sure if it's using them or defaulting them to certain values. The TTS quality of the Turbo version also seems to be less gooder (LOL), or lower quality, that the original version. There needs to be more clarity around this.

Yes that would be very nice to know how te parameters are used in the 2 versions (original & turbo). From what I saw playing with diferent turbo presets having the SAME temeperature parameter (and, according to the log that should be the only parameter influencing the out since exaggeration & cfg_weights are ignored) there are tone and expression variation that should not exist according to the logic that only temperature is taking into account. Besides this, if only temperature is acting why should exist so many presets in the ui interface that vary from children story narrator to overly dramatic chef if they have same temperature of 0.8...since they should produce the same output?.

[appsteve]

Hi, in the presets.yaml of the ui there are several configurations that are displayed in the turbo model page, all of them have among the parameters both cfg_weights and exageration but in the server log there is this warning: "chatterbox.tts_turbo: CFG, min_p and exaggeration are not supported by Turbo version and will be ignored". My questions are 1) Are these two parameters ignored in the case of turbo model usage? 2) Arethese configuration valid only for the original chatterbox model usage?
Thanks

Thanks for bringing this up! Yeah, it looks like the Turbo version is not showing the CFG and Exaggeration settings, so not sure if it's using them or defaulting them to certain values. The TTS quality of the Turbo version also seems to be less gooder (LOL), or lower quality, that the original version. There needs to be more clarity around this.

Yes that would be very nice to know how te parameters are used in the 2 versions (original & turbo). From what I saw playing with diferent turbo presets having the SAME temeperature parameter (and, according to the log that should be the only parameter influencing the out since exaggeration & cfg_weights are ignored) there are tone and expression variation that should not exist according to the logic that only temperature is taking into account. Besides this, if only temperature is acting why should exist so many presets in the ui interface that vary from children story narrator to overly dramatic chef if they have same temperature of 0.8...since they should produce the same output?.

Totally. The different presets must be setting all the parameters, including CFG/Temperature without exposing them in the GUI.

[credo99]

Hi, in the presets.yaml of the ui there are several configurations that are displayed in the turbo model page, all of them have among the parameters both cfg_weights and exageration but in the server log there is this warning: "chatterbox.tts_turbo: CFG, min_p and exaggeration are not supported by Turbo version and will be ignored". My questions are 1) Are these two parameters ignored in the case of turbo model usage? 2) Arethese configuration valid only for the original chatterbox model usage?
Thanks

Thanks for bringing this up! Yeah, it looks like the Turbo version is not showing the CFG and Exaggeration settings, so not sure if it's using them or defaulting them to certain values. The TTS quality of the Turbo version also seems to be less gooder (LOL), or lower quality, that the original version. There needs to be more clarity around this.

Well...it seems that the "culprit" is ResembleAI...check for line 299 in the turbo yaml file here https://huggingface.co/ResembleAI/chatterbox-turbo/blob/main/t3_turbo_v1.yaml supports_cfg: false .....nothing mentioned about exaggeration though