Velvet grew out of a need for a more accurate, free and open source voice analysis application that could also be trustworthy. This application is intended to assist trans feminine people in voice training by providing data driven feedback in real time. This application is free and always will be. It collects no data and does not communicate with any other services.
Velvet is a front end web application using React, Material UI and MeydaJS libraries. The application uses the web audio API for microphone access, audio filters and initial signal processing.
Velvet isolates the Fundamental Frequency and the First Formant Frequency from an audio stream. This information should give an accurate estimate of the range of frequencies in a human voice.
Currently a lowpass filter is applied at 300hz and a highpass filter at 30hz. This assists in filtering out background noise and sounds not in the standard vocal range. These settings may be updatable in a future version.
Audio files can be uploaded for analysis with a maximum size of 10mb. Files are not saved or retained in any way and must be uploaded each time.
Velvet is not currently capable of recording(saving) the audio signal. In the future the option for saving audio locally may be implemented. This application will not be hosting any data via a database.
Testing is ongoing with the signal processing algorithms and filters and will continue to be adjusted for usefulness and accurate feedback.
The current deployed version of Velvet can be found at https://velvet.rileyalexis.com.