Example training script:"Hello, my name is [Speaker Name]. I work at [Company Name] as a customer service representative. Today I'll be reading various sentences to help train an AI voice model. The weather is beautiful today with clear blue skies. Our company provides excellent customer support services."
1
Upload Audio
Upload 3-10 minutes of clear voice samples
2
Processing
Resemble processes and analyzes vocal characteristics
3
Training
AI model trains on voice patterns (~20-30 minutes)
4
Quality Check
Test generated voice with sample text
5
Deployment
Voice becomes available via API with unique UUID
Optimize Training Quality:✅ Do:
Use consistent microphone and environment
Speak at natural pace and volume
Include various sentence types (questions, statements, emotions)
Use Case: Web applications, content creation
Quality: High-fidelity audio for premium experiences
File Size: Larger files, better for non-real-time use
import asynciofrom resemble_ai import ResembleStreamingclass CustomerServiceTTS: def __init__(self): self.voice_uuid = "customer-service-voice-uuid" self.project_uuid = "your-project-uuid" async def handle_customer_inquiry(self, customer_message, inquiry_type): # Select appropriate voice based on inquiry type if inquiry_type == "complaint": response_tone = "empathetic" text = f"I understand your concern and I'm here to help resolve this issue for you." elif inquiry_type == "sales": response_tone = "enthusiastic" text = f"I'd be happy to tell you more about that product!" else: response_tone = "professional" text = f"Thank you for contacting us. How may I assist you today?" # Stream TTS response async for audio_chunk in self.stream_response(text): yield audio_chunk async def stream_response(self, text): # Implementation details for streaming pass
class BrandSpokesperson: def __init__(self, brand_voice_uuid): self.brand_voice = brand_voice_uuid self.brand_phrases = { "greeting": "Welcome to [Brand Name], where innovation meets excellence.", "closing": "Thank you for choosing [Brand Name]. We look forward to serving you.", "value_prop": "At [Brand Name], we believe in delivering exceptional value to every customer." } async def deliver_message(self, message_type, custom_content=None): if message_type in self.brand_phrases: text = self.brand_phrases[message_type] else: text = custom_content # Ensure brand consistency text = self.apply_brand_tone(text) return await self.synthesize_with_brand_voice(text) def apply_brand_tone(self, text): # Add brand-specific modifications # (tone adjustments, terminology, etc.) return text
class MultiVoiceSystem: def __init__(self): self.voices = { "announcer": "announcer-voice-uuid", "customer_service": "cs-voice-uuid", "technical_expert": "tech-voice-uuid", "sales_rep": "sales-voice-uuid" } async def route_to_appropriate_voice(self, message, context): # Determine which voice to use based on context if context.get("department") == "technical": voice_uuid = self.voices["technical_expert"] elif context.get("intent") == "purchase": voice_uuid = self.voices["sales_rep"] elif context.get("type") == "announcement": voice_uuid = self.voices["announcer"] else: voice_uuid = self.voices["customer_service"] return await self.synthesize_with_voice(message, voice_uuid)
Issue: Custom voice mispronounces specific wordsSolutions:
Include problematic words in training data
Use phonetic spelling in TTS requests
Create pronunciation guide for domain-specific terms
Retrain voice with additional samples if needed
Example Fix:
# Phonetic corrections for common issuesPRONUNCIATION_FIXES = { "API": "A P I", "HTTP": "H T T P", "OAuth": "O Auth", "UUID": "U U I D"}def apply_pronunciation_fixes(text): for term, phonetic in PRONUNCIATION_FIXES.items(): text = text.replace(term, phonetic) return text
Emotional Range Limitations
Issue: Voice sounds monotone or lacks expressionSolutions:
Include more emotional range in training samples
Use varied sentence types during training
Consider retraining with more expressive speaker
Test with TTS-specific emotional markup if available
Training Improvement:
Training script should include:- Questions: "How can I help you today?"- Excitement: "That's fantastic news!"- Concern: "I'm sorry to hear about that."- Professional: "Let me check that for you."
