OpenAI Launches GPT-Live-Transcribe and GPT-Transcribe Models via API
OpenAI has launched two new speech-to-text models via API: GPT-Live-Transcribe, designed for low-latency real-time streaming transcription, and GPT-Transcribe, optimized for asynchronous batch processing. These models offer improved context awareness and better handling of accents, noise, and technical jargon. These models significantly lower error rates compared to OpenAI's previous Whisper model, offering developers more accurate and cost-effective options for building voice-enabled applications. GPT-Live-Transcribe is priced at $0.017 per minute, while GPT-Transcribe costs $0.0045 per minute. In benchmarks, GPT-Transcribe achieved a 19.27% error rate on Common Voice across 22 languages, compared to Whisper's 40.37%.
## BACKGROUND
Automatic Speech Recognition (ASR) systems convert spoken audio into written text. Context-aware ASR improves this process by using surrounding text, prompts, or metadata to bias the model toward correctly transcribing rare words, proper nouns, and domain-specific jargon.