backend/services/ai/tts.js

import { AI_VOICE_PROMPT } from '../../../prompts/aiVoice.js';

import { tryModels, getPrompt, DEFAULT_SAFETY_SETTINGS } from '@/backend/services/ai/utils';
import { Modality } from '@google/genai';

export async function tts(text, voiceName, tone, apiKey, isOwnApi = false) {
    // Strict requirement: Max 3500 characters
    const ABSOLUTE_MAX_LENGTH = 3500;
    if (text && text.length > ABSOLUTE_MAX_LENGTH) {
        throw new Error(`Text is too long (${text.length} chars). Maximum allowed is ${ABSOLUTE_MAX_LENGTH} characters.`);
    }

    const models = ['gemini-2.5-flash-preview-tts', 'gemini-2.5-pro-preview-tts']; 
    const promptInstructions = AI_VOICE_PROMPT(tone);
    
    // For TTS models, prepending the tone instructions to the text is more stable than systemInstruction
    const textWithInstructions = `${promptInstructions}\n\nSCRIPT TO SPEAK:\n${text}`;

    console.log(`[TTS] Generating content... Length: ${text?.length || 0}`);

    const chunkBase64 = await tryModels(apiKey, models, async (ai, model) => {
        const response = await ai.models.generateContent({
            model: model,
            contents: [{ parts: [{ text: textWithInstructions }] }],
            config: { 
                responseModalities: [Modality.AUDIO], 
                safetySettings: DEFAULT_SAFETY_SETTINGS,
                speechConfig: { 
                    voiceConfig: { 
                        prebuiltVoiceConfig: { 
                            voiceName: voiceName || 'Zephyr' 
                        } 
                    } 
                } 
            }
        });

        const part = response.candidates?.[0]?.content?.parts.find(p => p.inlineData);
        if (part?.inlineData?.data) return part.inlineData.data;
        throw new Error("EMPTY_AUDIO_DATA_FROM_GEMINI");
    });

    return chunkBase64;
}