Public Beta of Cabby LLM Voice Generation Model
We are thrilled to announce a significant update to Cabby that introduces custom Cabby LLM voice generation model. This update allows users to personalize their flight experience by seamlessly integrating their unique voice or preferred accent into the simulation. The feature is available now in public beta, and we encourage users to try it out and provide feedback to help us improve the experience further.
Technical details for geeksβ
π€ Standard Voice Cloningβ
The standard voice cloning method allows users to replicate their voice in any language, that was used during model training process. This involves capturing the nuances of the user's voice through advanced machine learning algorithms. The software analyzes various vocal attributes, such as pitch, tone, and cadence, to create a synthetic voice that closely resembles the original speaker. This method is straightforward and efficient, enabling users to hear their own voice as the captain or crew member in the flight simulation, regardless of the language being spoken.
π€ Accent-Specific Voice Generation (Experimental)β
The second method focuses on generating voices with specific accents, which presents unique challenges. Unlike the standard voice cloning method, accent generation requires a different approach due to the lack of available LLM models and data that can effectively teach a model to produce multiple accents across various languages. To address this, Cabby employs a generic accent model that serves as a foundation for accent generation.
Once the base voice is established, the software applies a technique known as tone-coloring. This process involves adjusting the pitch and intonation of the voice to reflect the desired accent while maintaining the original vocal characteristics. By fine-tuning these elements, users can achieve a more authentic representation of their voice with the specific accent they wish to emulate.
Please keep in mind that output audio of the different models may differ in terms of quality and similarity.
How to Use Custom LLM Voice Generation?β
Go to the general settings in the Cabby app and scroll down until you see the switch shown below. Enable it.
From now on all the new voices added to the app will be generated using the new LLM model. If you want to use the old model - simply use previously added voices.
New voices are displayed with a special prefix [EXP]
to indicate that they were generated using the new model.
To use accents - simply select the voice you want to use in the pre-flight screen and select the accent you want to use.
Currently available accents:
- English (British)
- English (American)
- English (Australian)
- English (Indian)
When using your voice for the first time in a specific language, it may take a while to generate the voice. Please be patient. It should be much faster for subsequent uses.
Other changesβ
π Preview of the new voice generation modelβ
You can now preview the voice generated using the new model. Simply click "NEXT" in the voice creation screen and you will see the preview of the voice.
π Custom voice selection from fileβ
Now you don't have to rely on the computer's microphone to record your voice. You can simply record your voice using any device and save it as a file. Then you can upload it to the app and use it as a custom voice. Only .wav
files are supported.
π Other changesβ
- switch to descend phase if current one is "climb", vertical speed is less than -1500ft/min and the destination is close
- automatically show pre-flight settings page if the generation cannot be started due to missing settings
- setting for displaying generation progress on the pre-flight screen
Conclusionβ
With the introduction of these two methodsβstandard voice cloning and accent-specific voice generationβCabby is exploring new possibilities in flight simulation experiences. While these features are still experimental, they offer users the chance to personalize their flying adventures in exciting ways. We look forward to gathering feedback from our users as they experiment with these new capabilities, helping us refine and improve the technology for future updates. Your insights will be invaluable as we continue to enhance Cabbyβs offerings.
That's not all!
Cabby LLM is hosted on our dedicated servers, which means we now have full control over the generation process. This opens up new possibilities for future updates and improvements, such as enhanced voice quality, additional customization options, etc. We are committed to delivering the best possible experience to our users and look forward to sharing more exciting updates in the near future.
Supportβ
If you are happy with the service and appreciate our work, you can always support us by donating. It's not required, but it helps us keep the project alive and motivates us to work on new features and updates. Even a small donation can make a big difference!