Verified65% confidenceFactExact time
Early ChatGPT voice used a cascaded architecture: first converting speech to text, then having the large language model generate a response, and finally converting it to speech for output.
3
Sources
65%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
UnverifiedModels that truly natively support speech output (such as GPT-4o's real-time voice mode) adopt an approach of directly modeling speech tokens, skipping the intermediate text state, but such models are usually closed-source, difficult to deploy privately, and perform worse than pure-text reasoning models on complex tasks such as tool calling.77% similarUnverifiedCommonly used industry solutions include Twilio Voice API for call ingestion, models such as Whisper for transcription, and GPT-4o or similar models for intent understanding and response generation.76% similarUnverifiedThe new version's voice module exposes streaming transcription options, supports custom OpenAI TTS voice IDs, and forwards language and prompt parameters for streaming STT.76% similarVerifiedThe new generation of voice interaction has shifted to an end-to-end multimodal model architecture that understands and generates directly at the speech signal level, compressing latency and preserving paralinguistic information.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/516530API
curl https://kongchang.com/api/v1/knowledge/claims/516530MCP
get_claim(id=516530)