Why are you doing subscriptions?
While I am committed to maintaining a robust free tier, I also need to cover the cost of developing, maintaining and hopefully scaling Practy. If the demand is there, I would love to continue developing Practy full-time and building it as a larger platform with more content supported, exercise types and languages supported.
I personally feel that ads are distracting and harmful to the user experience, while not generating enough revenue to justify the trade-off.
To be transparent, AI is used in a few places throughout Practy. Audio files are generated using text-to-speech, and song recognition uses AI. LLMs are used for translation and definition creation. While AI is powerful, it can produce errors and slop. I take precautions to ensure AI outputs are:
•
Auditable and verifiable - Outputs are stored (not generated dynamically every time) so that users can flag errors and outputs can be reviewed
•
Structured - Instead of open-ended prompts, the LLM picks from pre-defined options (e.g. parts of speech, tags) to increase consistency and reduce errors
All that said, I do not view Practy as an AI-app and I do not believe AI is a replacement for human review and learning. Practy uses AI to help me as a solo developer due to limited resources.
Where does the audio come from?
Practy uses Google Cloud text-to-speech integration to generate audio files. Given enough demand, I would love to use either a higher quality TTS or even better, real native human recorded audio but that is not yet feasible.
What is spaced repetition?
Spaced repetition is a technique used to improve learning by repeating information at increasingly longer intervals as it is learned. This means you review information when you need to see it, rather than reviewing things you already know. Practy specifically uses the SuperMemo2 algorithm, similar to the algorithm used by Anki.
How does song recognition work?
Song recognition is provided in partnership with ACRCloud. For more information on ACRCloud, see the following description from their website:
“ACRCloud utilises patented Automatic Content Recognition (ACR) technology to enable the generation of a unique real-time fingerprint to identify in a matter of seconds, the content being played via an audio or video source which is typically a first screen, in order to trigger an action. With proprietary audio identification technology, ACRCloud’s is able to identify millions of hours of content in both a manageable and efficient manner that is highly relevant for advertisers, broadcasters, video streaming providers, music services, consumer electronics manufacturers and app developers. ACR’s fingerprint library includes 180 million+ audio tracks, which is one of the biggest in the world.”