Default Values:
• Model: GPT-5 Nano - Fastest and most cost-effective model, optimized for real-time conversations with low latency. Best for interactive tutoring where speed matters.
• Conversation History: 5 turns - Balances context awareness with token efficiency. Keeps recent conversation context while avoiding excessive token usage.
Custom prompt is saved in browser storage and will be used for all API calls. Enter only the base system prompt text. The backend automatically appends [CONVERSATION HISTORY], [NEW STUDENT INPUT], and [ERROR METADATA] sections.
⚙️ PA Configuration
Default Values:
• Critical Error Threshold: 65% - Words or phonemes scoring below this threshold are flagged as critical errors and included in the [ERROR METADATA] sent to the LLM. Lower values (e.g., 50%) are more lenient and flag fewer errors, while higher values (e.g., 80%) are stricter and flag more pronunciation issues. The default 65% provides a balanced approach that catches significant pronunciation problems without being overly sensitive.
The threshold is saved in browser storage and will be used for all pronunciation assessments. Critical errors help the tutor provide targeted feedback on specific pronunciation issues.
⚙️ TTS Configuration
Default Values:
• Model: TTS-1 - Optimized for real-time applications with fast generation and low latency. Best balance of speed and quality for interactive conversations.
• Voice: Sage - Clear, confident, and friendly female voice with professional American accent. Excellent for educational content, voice assistants, and customer support applications.
Voice Instructions: Control how the TTS model delivers the text. Only supported by GPT-4o Mini TTS model. For TTS-1 and TTS-1 HD, instructions are ignored. Leave empty to use default voice settings. Instructions are saved in browser storage. TTS playground: https://www.openai.fm/