Tag

speech tokenization

1 views collected around this technical thread.

Tencent Cloud Developer
Tencent Cloud Developer
Jun 14, 2024 · Artificial Intelligence

GPT-4o Speech Multimodal Technology: Speech Tokenization, LLM Integration, and Zero-shot TTS

GPT‑4o’s speech multimodal system discretizes audio into semantic and acoustic tokens, integrates these tokens with large language models through multi‑stage instruction tuning, and employs hierarchical zero‑shot text‑to‑speech decoding, enabling low‑latency, streaming, and prompt‑driven voice synthesis for applications like gaming.

AudioLMGPT-4oLLM integration
0 likes · 33 min read
GPT-4o Speech Multimodal Technology: Speech Tokenization, LLM Integration, and Zero-shot TTS