AI Disclosure
The Praxis Coach is an educational support tool, not a substitute for a qualified teacher. Coaching responses are generated by a large language model guided by your real-time pitch and session data. Feedback may occasionally be inaccurate.
What the Praxis Coach Is
The Praxis Coach (“Coach”) is a conversational assistant designed to support your saxophone practice. It uses on-device pitch detection data and a large language model (LLM) to generate real-time textual and voice feedback on your tone, rhythm, pitch accuracy, and articulation.
The Coach is powered by Google Vertex AI. See our Sub-processor Disclosure for details on how your data is handled.
What the Praxis Coach Is Not
The Coach is not a substitute for a qualified music instructor. It cannot replace the nuanced judgment, ear, and adaptive teaching ability of a human teacher. We recommend:
- Using the Coach alongside in-person or video lessons, not instead of them.
- Consulting a qualified teacher if you have concerns about posture, embouchure formation, or technique that could cause injury.
- Treating Coach recommendations as suggestions to explore, not authoritative directives.
How Coaching Recommendations Are Generated
When you play a note or phrase, the on-device pitch detection engine analyzes your audio in real time and produces structured performance data (pitch frequency, timing, amplitude, etc.). This data — combined with your session goals and Coach conversation history — is sent to the Praxis Coach backend (Google Vertex AI), which generates a natural-language response.
The LLM generating Coach responses was trained on large corpora of general text; it has not been trained on your personal data or your personal licks. Saxophone-specific guidance in the Coach is shaped by system-level instructions we provide, not by proprietary training on copyrighted method books or scores.
Accuracy and Limitations
- Pitch detection accuracy: Our on-device pitch detector is optimized for saxophone but may have reduced accuracy in very high overtone registers, with certain extended techniques (multiphonics, altissimo), or in noisy environments.
- Coach response quality: LLM responses may occasionally contain errors, misidentify the root cause of an issue, or give advice that does not apply to your specific situation.
- No real-time video: The Coach listens but cannot see your embouchure, posture, or hand position. It cannot flag issues that require visual assessment.
- Tempo and latency: Real-time feedback is limited by your device’s processing capability. On older devices, some feedback may be slightly delayed.
Feedback on Coach Quality
If you receive a Coach response that seems incorrect, unhelpful, or inappropriate, please report it via Settings → Feedback in the App. Your report helps us improve the Coach instructions and safety guardrails. Do not rely on any single Coach response as definitive guidance.
Backing Track Generation
Custom backing tracks are generated by Google Lyria (via Google Vertex AI) based on the tempo, key, and style you specify. These tracks are AI-generated instrumental accompaniments and do not reproduce copyrighted recordings.
Audio files you upload for stem separation (separating a recording into its instrument parts) are processed by LALAL.AI, which is a separate sub-processor.
See our Sub-processor Disclosure and our Terms of Service Section 7.
Data Not Used for Training
Your audio, personal licks, and practice sessions are not used to train AI models — either ClefScape’s or Google’s. See our Privacy Policy and Personal Licks Policy for details.
AI Quality Monitoring
To monitor the quality of Coach responses and improve the service, every Coach interaction sends a tracing record to Arize Phoenix, our AI observability sub-processor. By default this record contains only metadata — a one-way hashed user identifier (not reversible to your identity), session and request IDs, the prompt version we used, the tools called, your subscription tier, and your app version. Your raw Coach query text and the Coach’s response text are not sent to Phoenix in standard accounts.
Internal Praxis Saxophone accounts can opt in to raw-text logging for quality investigation; standard user accounts never have raw text sent. Your personal lick source attribution is excluded from Phoenix in all cases. See our Sub-processor Disclosure for details.
Questions about AI in Praxis Saxophone? Contact support@clefscape.com.